@fugood/buttress-server-poc 2.23.0-beta.32 → 2.23.0-beta.34

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
package/bin/start.mjs CHANGED
@@ -109,8 +109,77 @@ const {
109
109
  showSttModelsTable,
110
110
  startServer,
111
111
  checkAndNotifyUpdates,
112
+ startModelDownload,
112
113
  } = serverModule
113
114
 
115
+ /**
116
+ * Start pre-downloading models that have `download = true` in config.
117
+ * This allows models to be downloaded in the background at server startup
118
+ * so they are ready when clients request them.
119
+ *
120
+ * @param {Object} config - The parsed config object
121
+ */
122
+ async function startConfiguredDownloads(config) {
123
+ if (!config?.generators || !Array.isArray(config.generators)) return
124
+
125
+ // Filter generators that have download enabled and valid type
126
+ const downloadableGenerators = config.generators.filter((generator) => {
127
+ if (!generator.model?.download) return false
128
+ const { type } = generator
129
+ if (!type || (type !== 'ggml-llm' && type !== 'ggml-stt')) {
130
+ console.warn(`[Download] Skipping unknown generator type: ${type}`)
131
+ return false
132
+ }
133
+ return true
134
+ })
135
+
136
+ if (downloadableGenerators.length === 0) return
137
+
138
+ const { server: _server, generators: _generators, ...globalDefaults } = config
139
+
140
+ const downloadPromises = downloadableGenerators.map((generator) => {
141
+ const { type } = generator
142
+ const repoId = generator.model?.repo_id
143
+ console.log(`[Download] Starting pre-download for ${type}: ${repoId}`)
144
+
145
+ // Build config for this generator (merge global defaults)
146
+ const mergedConfig = {
147
+ ...globalDefaults,
148
+ backend: generator.backend || {},
149
+ model: generator.model || {},
150
+ runtime: { ...globalDefaults.runtime, ...(generator.runtime || {}) },
151
+ }
152
+
153
+ // Start download in background
154
+ return startModelDownload(type, mergedConfig, {
155
+ onProgress: () => {
156
+ // Progress is logged by the download function
157
+ },
158
+ onComplete: ({ repoId: modelId, alreadyExists }) => {
159
+ if (alreadyExists) {
160
+ console.log(`[Download] Pre-download complete (already exists): ${modelId}`)
161
+ } else {
162
+ console.log(`[Download] Pre-download complete: ${modelId}`)
163
+ }
164
+ },
165
+ onError: (error) => {
166
+ console.error(`[Download] Pre-download failed for ${repoId}:`, error.message)
167
+ },
168
+ })
169
+ })
170
+
171
+ // Wait for all initial download checks to complete
172
+ // (the actual downloads continue in the background)
173
+ const results = await Promise.all(downloadPromises)
174
+ const started = results.filter((r) => r.started).length
175
+ const existing = results.filter((r) => r.alreadyExists).length
176
+ const downloading = results.filter((r) => r.alreadyDownloading).length
177
+
178
+ console.log(
179
+ `[Download] Pre-download summary: ${started} started, ${existing} already exist, ${downloading} already downloading`,
180
+ )
181
+ }
182
+
114
183
  // Default models for --test-models-default
115
184
  const DEFAULT_TEST_GGML_LLM_MODELS = [
116
185
  'ggml-org/gpt-oss-20b-GGUF',
@@ -263,6 +332,11 @@ startServer({
263
332
  console.log(' Set ENABLE_OPENAI_COMPAT_ENDPOINT=1 to enable')
264
333
  console.log()
265
334
  }
335
+
336
+ // Start pre-downloading models with download = true
337
+ if (defaultConfig) {
338
+ await startConfiguredDownloads(defaultConfig)
339
+ }
266
340
  })
267
341
  .catch((error) => {
268
342
  console.error('Failed to start Buttress server POC:', error)
@@ -15,9 +15,9 @@ cache_dir = "./.buttress-cache"
15
15
  # huggingface_token = "hf_xx"
16
16
 
17
17
  # Global model params
18
- flash_attn_type = "on"
19
- cache_type_k = "q8_0"
20
- cache_type_v = "q8_0"
18
+ flash_attn_type = "off"
19
+ cache_type_k = "f16"
20
+ cache_type_v = "f16"
21
21
  cpu_mask = "0xfc"
22
22
  cpu_strict = true
23
23
  n_threads = 6
@@ -39,6 +39,7 @@ cpu_memory_fraction = 0.8
39
39
  [generators.model]
40
40
  repo_id = "ggml-org/gpt-oss-20b-GGUF"
41
41
  quantization = "mxfp4"
42
+ download = true
42
43
  n_ctx = 51200
43
44
 
44
45
  [[generators]]
@@ -51,6 +52,7 @@ cpu_memory_fraction = 0.8
51
52
  [generators.model]
52
53
  repo_id = "unsloth/Qwen3-30B-A3B-Instruct-2507-GGUF"
53
54
  quantization = "q4_0"
55
+ download = true
54
56
  n_ctx = 51200
55
57
 
56
58
  [[generators]]
@@ -63,6 +65,7 @@ cpu_memory_fraction = 0.8
63
65
  [generators.model]
64
66
  repo_id = "ggml-org/gemma-3-12b-it-qat-GGUF"
65
67
  quantization = "q4_0"
68
+ download = true
66
69
  n_ctx = 51200
67
70
 
68
71
  # Speech-to-Text (STT) generators
@@ -74,4 +77,5 @@ variant_preference = ["default"]
74
77
  [generators.model]
75
78
  repo_id = "BricksDisplay/whisper-ggml"
76
79
  filename = "ggml-base.bin"
80
+ download = true
77
81
  use_gpu = false
@@ -1,4 +1,10 @@
1
1
  # Sample Buttress server configuration with per-model defaults
2
+ #
3
+ # Model Pre-Download:
4
+ # Set `download = true` on a generator's [generators.model] section to
5
+ # automatically download the model at server startup if it doesn't exist.
6
+ # When initContext is called, it will wait for any in-progress download
7
+ # of the same model before proceeding.
2
8
 
3
9
  # Environment variables (only set if not already defined)
4
10
  [env]
@@ -10,6 +16,10 @@ port = 2080
10
16
  log_level = "info"
11
17
  # max_body_size = "100MB" # Supports string (e.g., "100MB", "1GB") or number in bytes
12
18
 
19
+ [openai_compat]
20
+ # cors_allowed_origins = ["http://localhost:3000", "https://example.com"] # Restrict to specific origins
21
+ # cors_allowed_origins = "*" # Allow all origins (default)
22
+
13
23
  [runtime]
14
24
  cache_dir = "./.buttress-cache"
15
25
  # huggingface_token = "hf_xx"
@@ -19,6 +29,10 @@ flash_attn_type = "on"
19
29
  cache_type_k = "q8_0"
20
30
  cache_type_v = "q8_0"
21
31
 
32
+ # Delay before releasing context when refCount reaches 0 (default: 10000ms)
33
+ # Set to 0 for immediate release
34
+ context_release_delay_ms = 30000
35
+
22
36
  # Session state cache for ggml-llm (saves KV cache to disk for prompt reuse)
23
37
  [runtime.session_cache]
24
38
  enabled = true
@@ -34,6 +48,7 @@ cpu_memory_fraction = 0.95
34
48
  [generators.model]
35
49
  repo_id = "ggml-org/gpt-oss-20b-GGUF"
36
50
  quantization = "mxfp4"
51
+ download = true
37
52
  # n_ctx = 12800 # Max: 131072
38
53
 
39
54
  [[generators]]
@@ -44,6 +59,17 @@ gpu_memory_fraction = 0.95
44
59
  [generators.model]
45
60
  repo_id = "ggml-org/gpt-oss-120b-GGUF"
46
61
  quantization = "mxfp4"
62
+ download = false
63
+
64
+ [[generators]]
65
+ type = "ggml-llm"
66
+ [generators.backend]
67
+ variant_preference = ["cuda", "vulkan", "snapdragon", "default"]
68
+ gpu_memory_fraction = 0.95
69
+ [generators.model]
70
+ repo_id = "unsloth/Qwen3-30B-A3B-Instruct-2507-GGUF"
71
+ quantization = "q4_0"
72
+ download = true
47
73
 
48
74
  [[generators]]
49
75
  type = "ggml-llm"
@@ -65,6 +91,7 @@ gpu_memory_fraction = 0.95
65
91
  [generators.model]
66
92
  repo_id = "bartowski/mistralai_Devstral-2-123B-Instruct-2512-GGUF"
67
93
  quantization = "q4_0"
94
+ download = false
68
95
  # n_ctx = 128000 # Max: 262144
69
96
 
70
97
  [[generators]]
@@ -74,6 +101,7 @@ variant_preference = ["default"]
74
101
  [generators.model]
75
102
  repo_id = "ggml-org/gemma-3-270m-qat-GGUF"
76
103
  quantization = "q4_0"
104
+ download = true
77
105
 
78
106
  [[generators]]
79
107
  type = "ggml-llm"
@@ -81,6 +109,7 @@ type = "ggml-llm"
81
109
  variant_preference = ["default"]
82
110
  [generators.model]
83
111
  repo_id = "ggml-org/gemma-3-12b-it-qat-GGUF"
112
+ download = false
84
113
 
85
114
  # Speech-to-Text (STT) generators
86
115
  [[generators]]
@@ -90,6 +119,7 @@ variant_preference = ["cuda", "vulkan", "default"]
90
119
  [generators.model]
91
120
  repo_id = "BricksDisplay/whisper-ggml"
92
121
  filename = "ggml-small-q8_0.bin"
122
+ download = true
93
123
  use_gpu = true
94
124
 
95
125
  [[generators]]
@@ -99,5 +129,6 @@ variant_preference = ["cuda", "vulkan", "default"]
99
129
  [generators.model]
100
130
  repo_id = "BricksDisplay/whisper-ggml"
101
131
  filename = "ggml-large-v3-turbo-q8_0.bin"
132
+ download = false
102
133
  use_gpu = true
103
134
  use_flash_attn = true
package/lib/index.js CHANGED
@@ -1 +1 @@
1
- import{createRequire as e}from"node:module";const t=e(import.meta.url);import{initTRPC as n}from"@trpc/server";import{z as r}from"zod";import{gguf as o}from"@huggingface/gguf";import{getBackendDevicesInfo as a,isLibVariantAvailable as i,loadModel as s}from"@fugood/llama.node";import{initWhisper as l}from"@fugood/whisper.node";import{default as c}from"@iarna/toml";var u,d,p={285:e=>{e.exports=function(e,t){if("string"==typeof e)return i(e);if("number"==typeof e)return a(e,t);return null},e.exports.format=a,e.exports.parse=i;var t=/\B(?=(\d{3})+(?!\d))/g,n=/(?:\.0*|(\.[^0]+)0+)$/,r={b:1,kb:1024,mb:1<<20,gb:1<<30,tb:Math.pow(1024,4),pb:Math.pow(1024,5)},o=/^((-|\+)?(\d+(?:\.\d+)?)) *(kb|mb|gb|tb|pb)$/i;function a(e,o){if(!Number.isFinite(e))return null;var a=Math.abs(e),i=o&&o.thousandsSeparator||"",s=o&&o.unitSeparator||"",l=o&&void 0!==o.decimalPlaces?o.decimalPlaces:2,c=Boolean(o&&o.fixedDecimals),u=o&&o.unit||"";u&&r[u.toLowerCase()]||(u=a>=r.pb?"PB":a>=r.tb?"TB":a>=r.gb?"GB":a>=r.mb?"MB":a>=r.kb?"KB":"B");var d=(e/r[u.toLowerCase()]).toFixed(l);return c||(d=d.replace(n,"$1")),i&&(d=d.split(".").map(function(e,n){return 0===n?e.replace(t,i):e}).join(".")),d+s+u}function i(e){if("number"==typeof e&&!isNaN(e))return e;if("string"!=typeof e)return null;var t,n=o.exec(e),a="b";return n?(t=parseFloat(n[1]),a=n[4].toLowerCase()):(t=parseInt(e,10),a="b"),isNaN(t)?null:Math.floor(r[a]*t)}}},m={};function f(e){var t=m[e];if(void 0!==t)return t.exports;var n=m[e]={exports:{}};return p[e](n,n.exports,f),n.exports}d=Object.getPrototypeOf?e=>Object.getPrototypeOf(e):e=>e.__proto__,f.t=function(e,t){if(1&t&&(e=this(e)),8&t)return e;if("object"==typeof e&&e){if(4&t&&e.__esModule)return e;if(16&t&&"function"==typeof e.then)return e}var n=Object.create(null);f.r(n);var r={};u=u||[null,d({}),d([]),d(d)];for(var o=2&t&&e;("object"==typeof o||"function"==typeof o)&&!~u.indexOf(o);o=d(o))Object.getOwnPropertyNames(o).forEach(t=>r[t]=()=>e[t]);return r.default=()=>e,f.d(n,r),n},f.d=(e,t)=>{for(var n in t)f.o(t,n)&&!f.o(e,n)&&Object.defineProperty(e,n,{enumerable:!0,get:t[n]})},f.o=(e,t)=>Object.prototype.hasOwnProperty.call(e,t),f.r=e=>{"undefined"!=typeof Symbol&&Symbol.toStringTag&&Object.defineProperty(e,Symbol.toStringTag,{value:"Module"}),Object.defineProperty(e,"__esModule",{value:!0})};var h={};f.r(h),f.d(h,{finalizeGenerator:()=>to,generatorRegistry:()=>Zr,getCapabilities:()=>Lr,getModelIdentifier:()=>oo,ggmlLlm:()=>no,ggmlStt:()=>ro,showModelsTable:()=>Kr,showSttModelsTable:()=>Wr,startGenerator:()=>eo,status:()=>ao,testGgmlLlmCapabilities:()=>Yr,testGgmlSttCapabilities:()=>Qr});const y=t("node:fs/promises"),g=t("node:path"),b=t("node:url"),v={PARSE_ERROR:-32700,BAD_REQUEST:-32600,INTERNAL_SERVER_ERROR:-32603,NOT_IMPLEMENTED:-32603,BAD_GATEWAY:-32603,SERVICE_UNAVAILABLE:-32603,GATEWAY_TIMEOUT:-32603,UNAUTHORIZED:-32001,PAYMENT_REQUIRED:-32002,FORBIDDEN:-32003,NOT_FOUND:-32004,METHOD_NOT_SUPPORTED:-32005,TIMEOUT:-32008,CONFLICT:-32009,PRECONDITION_FAILED:-32012,PAYLOAD_TOO_LARGE:-32013,UNSUPPORTED_MEDIA_TYPE:-32015,UNPROCESSABLE_CONTENT:-32022,PRECONDITION_REQUIRED:-32028,TOO_MANY_REQUESTS:-32029,CLIENT_CLOSED_REQUEST:-32099},w={[-32700]:"PARSE_ERROR",[-32600]:"BAD_REQUEST",[-32603]:"INTERNAL_SERVER_ERROR",[-32001]:"UNAUTHORIZED",[-32002]:"PAYMENT_REQUIRED",[-32003]:"FORBIDDEN",[-32004]:"NOT_FOUND",[-32005]:"METHOD_NOT_SUPPORTED",[-32008]:"TIMEOUT",[-32009]:"CONFLICT",[-32012]:"PRECONDITION_FAILED",[-32013]:"PAYLOAD_TOO_LARGE",[-32015]:"UNSUPPORTED_MEDIA_TYPE",[-32022]:"UNPROCESSABLE_CONTENT",[-32028]:"PRECONDITION_REQUIRED",[-32029]:"TOO_MANY_REQUESTS",[-32099]:"CLIENT_CLOSED_REQUEST"};function _(e){return!!e&&!Array.isArray(e)&&"object"==typeof e}const x="function"==typeof Symbol&&!!Symbol.asyncIterator;function S(e){return x&&_(e)&&Symbol.asyncIterator in e}const k=e=>e();function C(e){return e}var E=Object.create,M=Object.defineProperty,T=Object.getOwnPropertyDescriptor,B=Object.getOwnPropertyNames,$=Object.getPrototypeOf,O=Object.prototype.hasOwnProperty,N=(e,t)=>function(){return t||(0,e[B(e)[0]])((t={exports:{}}).exports,t),t.exports},I=(e,t,n)=>(n=null!=e?E($(e)):{},((e,t,n,r)=>{if(t&&"object"==typeof t||"function"==typeof t)for(var o,a=B(t),i=0,s=a.length;i<s;i++)o=a[i],O.call(e,o)||o===n||M(e,o,{get:(e=>t[e]).bind(null,o),enumerable:!(r=T(t,o))||r.enumerable});return e})(!t&&e&&e.__esModule?n:M(n,"default",{value:e,enumerable:!0}),e));const P={PARSE_ERROR:400,BAD_REQUEST:400,UNAUTHORIZED:401,PAYMENT_REQUIRED:402,FORBIDDEN:403,NOT_FOUND:404,METHOD_NOT_SUPPORTED:405,TIMEOUT:408,CONFLICT:409,PRECONDITION_FAILED:412,PAYLOAD_TOO_LARGE:413,UNSUPPORTED_MEDIA_TYPE:415,UNPROCESSABLE_CONTENT:422,PRECONDITION_REQUIRED:428,TOO_MANY_REQUESTS:429,CLIENT_CLOSED_REQUEST:499,INTERNAL_SERVER_ERROR:500,NOT_IMPLEMENTED:501,BAD_GATEWAY:502,SERVICE_UNAVAILABLE:503,GATEWAY_TIMEOUT:504};function q(e){var t;return null!==(t=P[e])&&void 0!==t?t:500}function A(e){return q(e.code)}var z=N({"../../node_modules/.pnpm/@oxc-project+runtime@0.72.2/node_modules/@oxc-project/runtime/src/helpers/typeof.js"(e,t){function n(e){return t.exports=n="function"==typeof Symbol&&"symbol"==typeof Symbol.iterator?function(e){return typeof e}:function(e){return e&&"function"==typeof Symbol&&e.constructor===Symbol&&e!==Symbol.prototype?"symbol":typeof e},t.exports.__esModule=!0,t.exports.default=t.exports,n(e)}t.exports=n,t.exports.__esModule=!0,t.exports.default=t.exports}}),F=N({"../../node_modules/.pnpm/@oxc-project+runtime@0.72.2/node_modules/@oxc-project/runtime/src/helpers/toPrimitive.js"(e,t){var n=z().default;t.exports=function(e,t){if("object"!=n(e)||!e)return e;var r=e[Symbol.toPrimitive];if(void 0!==r){var o=r.call(e,t||"default");if("object"!=n(o))return o;throw new TypeError("@@toPrimitive must return a primitive value.")}return("string"===t?String:Number)(e)},t.exports.__esModule=!0,t.exports.default=t.exports}}),U=N({"../../node_modules/.pnpm/@oxc-project+runtime@0.72.2/node_modules/@oxc-project/runtime/src/helpers/toPropertyKey.js"(e,t){var n=z().default,r=F();t.exports=function(e){var t=r(e,"string");return"symbol"==n(t)?t:t+""},t.exports.__esModule=!0,t.exports.default=t.exports}}),j=N({"../../node_modules/.pnpm/@oxc-project+runtime@0.72.2/node_modules/@oxc-project/runtime/src/helpers/defineProperty.js"(e,t){var n=U();t.exports=function(e,t,r){return(t=n(t))in e?Object.defineProperty(e,t,{value:r,enumerable:!0,configurable:!0,writable:!0}):e[t]=r,e},t.exports.__esModule=!0,t.exports.default=t.exports}}),D=N({"../../node_modules/.pnpm/@oxc-project+runtime@0.72.2/node_modules/@oxc-project/runtime/src/helpers/objectSpread2.js"(e,t){var n=j();function r(e,t){var n=Object.keys(e);if(Object.getOwnPropertySymbols){var r=Object.getOwnPropertySymbols(e);t&&(r=r.filter(function(t){return Object.getOwnPropertyDescriptor(e,t).enumerable})),n.push.apply(n,r)}return n}t.exports=function(e){for(var t=1;t<arguments.length;t++){var o=null!=arguments[t]?arguments[t]:{};t%2?r(Object(o),!0).forEach(function(t){n(e,t,o[t])}):Object.getOwnPropertyDescriptors?Object.defineProperties(e,Object.getOwnPropertyDescriptors(o)):r(Object(o)).forEach(function(t){Object.defineProperty(e,t,Object.getOwnPropertyDescriptor(o,t))})}return e},t.exports.__esModule=!0,t.exports.default=t.exports}}),R=I(D(),1);function L(e){const{path:t,error:n,config:r}=e,{code:o}=e.error,a={message:n.message,code:v[o],data:{code:o,httpStatus:A(n)}};return r.isDev&&"string"==typeof e.error.stack&&(a.data.stack=e.error.stack),"string"==typeof t&&(a.data.path=t),r.errorFormatter((0,R.default)((0,R.default)({},e),{},{shape:a}))}var G=I(j(),1),H=class extends Error{};function J(e){if(e instanceof V)return e;if(e instanceof Error&&"TRPCError"===e.name)return e;const t=new V({code:"INTERNAL_SERVER_ERROR",cause:e});return e instanceof Error&&e.stack&&(t.stack=e.stack),t}var V=class extends Error{constructor(e){var t,n,r;const o=function(e){if(e instanceof Error)return e;const t=typeof e;return"undefined"!==t&&"function"!==t&&null!==e?"object"!==t?new Error(String(e)):_(e)?Object.assign(new H,e):void 0:void 0}(e.cause);super(null!==(t=null!==(n=e.message)&&void 0!==n?n:null==o?void 0:o.message)&&void 0!==t?t:e.code,{cause:o}),(0,G.default)(this,"cause",void 0),(0,G.default)(this,"code",void 0),this.code=e.code,this.name="TRPCError",null!==(r=this.cause)&&void 0!==r||(this.cause=o)}},K=I(D(),1);function Y(e,t){return"error"in t?(0,K.default)((0,K.default)({},t),{},{error:e.transformer.output.serialize(t.error)}):"data"in t.result?(0,K.default)((0,K.default)({},t),{},{result:(0,K.default)((0,K.default)({},t.result),{},{data:e.transformer.output.serialize(t.result.data)})}):t}function W(e,t){return Array.isArray(t)?t.map(t=>Y(e,t)):Y(e,t)}Error;I(D(),1);async function Q(e,t){const{_def:n}=e;let r=n.procedures[t];for(;!r;){const e=Object.keys(n.lazy).find(e=>t.startsWith(e));if(!e)return null;const o=n.lazy[e];await o.load(),r=n.procedures[t]}return r}const Z=Symbol();function X(e){return Array.isArray(e)&&e[2]===Z}function ee(e){return"object"==typeof e&&null!==e&&"subscribe"in e}function te(e){const t={subscribe(t){let n=null,r=!1,o=!1,a=!1;function i(){null!==n?o||(o=!0,"function"==typeof n?n():n&&n.unsubscribe()):a=!0}return n=e({next(e){var n;r||null===(n=t.next)||void 0===n||n.call(t,e)},error(e){var n;r||(r=!0,null===(n=t.error)||void 0===n||n.call(t,e),i())},complete(){var e;r||(r=!0,null===(e=t.complete)||void 0===e||e.call(t),i())}}),a&&i(),{unsubscribe:i}},pipe:(...e)=>e.reduce(ne,t)};return t}function ne(e,t){return t(e)}function re(e,t){const n=function(e,t){let n=null;const r=()=>{null==n||n.unsubscribe(),n=null,t.removeEventListener("abort",r)};return new ReadableStream({start(o){n=e.subscribe({next(e){o.enqueue({ok:!0,value:e})},error(e){o.enqueue({ok:!1,error:e}),o.close()},complete(){o.close()}}),t.aborted?r():t.addEventListener("abort",r,{once:!0})},cancel(){r()}})}(e,t),r=n.getReader(),o={async next(){const e=await r.read();if(e.done)return{value:void 0,done:!0};const{value:t}=e;if(!t.ok)throw t.error;return{value:t.value,done:!1}},return:async()=>(await r.cancel(),{value:void 0,done:!0})};return{[Symbol.asyncIterator]:()=>o}}function oe(e){let t;try{t=JSON.parse(e)}catch(e){throw new V({code:"PARSE_ERROR",message:"Not JSON-parsable query params",cause:e})}return function(e){try{if(null===e)return null;if(!_(e))throw new Error("Expected object");const t=Object.entries(e).filter(([e,t])=>"string"!=typeof t);if(t.length>0)throw new Error(`Expected connectionParams to be string values. Got ${t.map(([e,t])=>`${e}: ${typeof t}`).join(", ")}`);return e}catch(e){throw new V({code:"PARSE_ERROR",message:"Invalid connection params shape",cause:e})}}(t)}var ae=I(D(),1);function ie(e){let t=null;const n=Symbol.for("@trpc/server/http/memo");let r=n;return{read:async()=>{var o;return r!==n||(null!==(o=t)&&void 0!==o||(t=e().catch(e=>{if(e instanceof V)throw e;throw new V({code:"BAD_REQUEST",message:e instanceof Error?e.message:"Invalid input",cause:e})})),r=await t,t=null),r},result:()=>r!==n?r:void 0}}const se={isMatch(e){var t;return!!(null===(t=e.headers.get("content-type"))||void 0===t?void 0:t.startsWith("application/json"))},async parse(e){var t;const{req:n}=e,r="1"===e.searchParams.get("batch"),o=r?e.path.split(","):[e.path],a=ie(async()=>{let t;if("GET"===n.method){const n=e.searchParams.get("input");n&&(t=JSON.parse(n))}else t=await n.json();if(void 0===t)return{};if(!r)return{0:e.router._def._config.transformer.input.deserialize(t)};if(!_(t))throw new V({code:"BAD_REQUEST",message:'"input" needs to be an object when doing a batch call'});const a={};for(const n of o.keys()){const r=t[n];void 0!==r&&(a[n]=e.router._def._config.transformer.input.deserialize(r))}return a}),i=await Promise.all(o.map(async(t,n)=>{const r=await Q(e.router,t);return{path:t,procedure:r,getRawInput:async()=>{let t=(await a.read())[n];if("subscription"===(null==r?void 0:r._def.type)){var o,i;const n=null!==(o=null!==(i=e.headers.get("last-event-id"))&&void 0!==i?i:e.searchParams.get("lastEventId"))&&void 0!==o?o:e.searchParams.get("Last-Event-Id");if(n)if(_(t))t=(0,ae.default)((0,ae.default)({},t),{},{lastEventId:n});else{var s;null!==(s=t)&&void 0!==s||(t={lastEventId:n})}}return t},result:()=>{var e;return null===(e=a.result())||void 0===e?void 0:e[n]}}})),s=new Set(i.map(e=>{var t;return null===(t=e.procedure)||void 0===t?void 0:t._def.type}).filter(Boolean));if(s.size>1)throw new V({code:"BAD_REQUEST",message:`Cannot mix procedure types in call: ${Array.from(s).join(", ")}`});const l=null!==(t=s.values().next().value)&&void 0!==t?t:"unknown",c=e.searchParams.get("connectionParams");return{isBatchCall:r,accept:n.headers.get("trpc-accept"),calls:i,type:l,connectionParams:null===c?null:oe(c),signal:n.signal,url:e.url}}},le=[se,{isMatch(e){var t;return!!(null===(t=e.headers.get("content-type"))||void 0===t?void 0:t.startsWith("multipart/form-data"))},async parse(e){const{req:t}=e;if("POST"!==t.method)throw new V({code:"METHOD_NOT_SUPPORTED",message:"Only POST requests are supported for multipart/form-data requests"});const n=ie(async()=>await t.formData()),r=await Q(e.router,e.path);return{accept:null,calls:[{path:e.path,getRawInput:n.read,result:n.result,procedure:r}],isBatchCall:!1,type:"mutation",connectionParams:null,signal:t.signal,url:e.url}}},{isMatch(e){var t;return!!(null===(t=e.headers.get("content-type"))||void 0===t?void 0:t.startsWith("application/octet-stream"))},async parse(e){const{req:t}=e;if("POST"!==t.method)throw new V({code:"METHOD_NOT_SUPPORTED",message:"Only POST requests are supported for application/octet-stream requests"});const n=ie(async()=>t.body);return{calls:[{path:e.path,getRawInput:n.read,result:n.result,procedure:await Q(e.router,e.path)}],isBatchCall:!1,accept:null,type:"mutation",connectionParams:null,signal:t.signal,url:e.url}}}];async function ce(e){const t=function(e){const t=le.find(t=>t.isMatch(e));if(t)return t;if(!t&&"GET"===e.method)return se;throw new V({code:"UNSUPPORTED_MEDIA_TYPE",message:e.headers.has("content-type")?`Unsupported content-type "${e.headers.get("content-type")}`:"Missing content-type header"})}(e.req);return await t.parse(e)}function ue(e){return _(e)&&"AbortError"===e.name}function de(e="AbortError"){throw new DOMException(e,"AbortError")}function pe(e){return"[object Object]"===Object.prototype.toString.call(e)}var me=I(j(),1);let fe;const he=new WeakMap,ye=()=>{};fe=Symbol.toStringTag;var ge,be,ve,we,_e=class e{constructor(e){(0,me.default)(this,"promise",void 0),(0,me.default)(this,"subscribers",[]),(0,me.default)(this,"settlement",null),(0,me.default)(this,fe,"Unpromise"),this.promise="function"==typeof e?new Promise(e):e;const t=this.promise.then(e=>{const{subscribers:t}=this;this.subscribers=null,this.settlement={status:"fulfilled",value:e},null==t||t.forEach(({resolve:t})=>{t(e)})});"catch"in t&&t.catch(e=>{const{subscribers:t}=this;this.subscribers=null,this.settlement={status:"rejected",reason:e},null==t||t.forEach(({reject:t})=>{t(e)})})}subscribe(){let e,t;const{settlement:n}=this;if(null===n){if(null===this.subscribers)throw new Error("Unpromise settled but still has subscribers");const n=function(){let e,t;const n=new Promise((n,r)=>{e=n,t=r});return{promise:n,resolve:e,reject:t}}();this.subscribers=(r=this.subscribers,o=n,[...r,o]),e=n.promise,t=()=>{null!==this.subscribers&&(this.subscribers=function(e,t){const n=e.indexOf(t);return-1!==n?function(e,t){return[...e.slice(0,t),...e.slice(t+1)]}(e,n):e}(this.subscribers,n))}}else{const{status:r}=n;e="fulfilled"===r?Promise.resolve(n.value):Promise.reject(n.reason),t=ye}var r,o;return Object.assign(e,{unsubscribe:t})}then(e,t){const n=this.subscribe(),{unsubscribe:r}=n;return Object.assign(n.then(e,t),{unsubscribe:r})}catch(e){const t=this.subscribe(),{unsubscribe:n}=t;return Object.assign(t.catch(e),{unsubscribe:n})}finally(e){const t=this.subscribe(),{unsubscribe:n}=t;return Object.assign(t.finally(e),{unsubscribe:n})}static proxy(t){const n=e.getSubscribablePromise(t);return void 0!==n?n:e.createSubscribablePromise(t)}static createSubscribablePromise(t){const n=new e(t);return he.set(t,n),he.set(n,n),n}static getSubscribablePromise(e){return he.get(e)}static resolve(t){const n="object"==typeof t&&null!==t&&"then"in t&&"function"==typeof t.then?t:Promise.resolve(t);return e.proxy(n).subscribe()}static async any(t){const n=(Array.isArray(t)?t:[...t]).map(e.resolve);try{return await Promise.any(n)}finally{n.forEach(({unsubscribe:e})=>{e()})}}static async race(t){const n=(Array.isArray(t)?t:[...t]).map(e.resolve);try{return await Promise.race(n)}finally{n.forEach(({unsubscribe:e})=>{e()})}}static async raceReferences(e){const t=e.map(xe);try{return await Promise.race(t)}finally{for(const e of t)e.unsubscribe()}}};function xe(e){return _e.proxy(e).then(()=>[e])}function Se(e,t){const n=e,r=n[Symbol.dispose];return n[Symbol.dispose]=()=>{t(),null==r||r()},n}function ke(e,t){const n=e,r=n[Symbol.asyncDispose];return n[Symbol.asyncDispose]=async()=>{await t(),await(null==r?void 0:r())},n}null!==(be=(ge=Symbol).dispose)&&void 0!==be||(ge.dispose=Symbol()),null!==(we=(ve=Symbol).asyncDispose)&&void 0!==we||(ve.asyncDispose=Symbol());const Ce=Symbol();function Ee(e){let t=null;return Se({start(){if(t)throw new Error("Timer already started");return new Promise(n=>{t=setTimeout(()=>n(Ce),e)})}},()=>{t&&clearTimeout(t)})}var Me=N({"../../node_modules/.pnpm/@oxc-project+runtime@0.72.2/node_modules/@oxc-project/runtime/src/helpers/usingCtx.js"(e,t){t.exports=function(){var e="function"==typeof SuppressedError?SuppressedError:function(e,t){var n=Error();return n.name="SuppressedError",n.error=e,n.suppressed=t,n},t={},n=[];function r(e,t){if(null!=t){if(Object(t)!==t)throw new TypeError("using declarations can only be used with objects, functions, null, or undefined.");if(e)var r=t[Symbol.asyncDispose||Symbol.for("Symbol.asyncDispose")];if(void 0===r&&(r=t[Symbol.dispose||Symbol.for("Symbol.dispose")],e))var o=r;if("function"!=typeof r)throw new TypeError("Object is not disposable.");o&&(r=function(){try{o.call(t)}catch(e){return Promise.reject(e)}}),n.push({v:t,d:r,a:e})}else e&&n.push({d:t,a:e});return t}return{e:t,u:r.bind(null,!1),a:r.bind(null,!0),d:function(){var r,o=this.e,a=0;function i(){for(;r=n.pop();)try{if(!r.a&&1===a)return a=0,n.push(r),Promise.resolve().then(i);if(r.d){var e=r.d.call(r.v);if(r.a)return a|=2,Promise.resolve(e).then(i,s)}else a|=1}catch(e){return s(e)}if(1===a)return o!==t?Promise.reject(o):Promise.resolve();if(o!==t)throw o}function s(n){return o=o!==t?new e(n,o):n,i()}return i()}}},t.exports.__esModule=!0,t.exports.default=t.exports}}),Te=N({"../../node_modules/.pnpm/@oxc-project+runtime@0.72.2/node_modules/@oxc-project/runtime/src/helpers/OverloadYield.js"(e,t){t.exports=function(e,t){this.v=e,this.k=t},t.exports.__esModule=!0,t.exports.default=t.exports}}),Be=N({"../../node_modules/.pnpm/@oxc-project+runtime@0.72.2/node_modules/@oxc-project/runtime/src/helpers/awaitAsyncGenerator.js"(e,t){var n=Te();t.exports=function(e){return new n(e,0)},t.exports.__esModule=!0,t.exports.default=t.exports}}),$e=N({"../../node_modules/.pnpm/@oxc-project+runtime@0.72.2/node_modules/@oxc-project/runtime/src/helpers/wrapAsyncGenerator.js"(e,t){var n=Te();function r(e){var t,r;function o(t,r){try{var i=e[t](r),s=i.value,l=s instanceof n;Promise.resolve(l?s.v:s).then(function(n){if(l){var r="return"===t?"return":"next";if(!s.k||n.done)return o(r,n);n=e[r](n).value}a(i.done?"return":"normal",n)},function(e){o("throw",e)})}catch(e){a("throw",e)}}function a(e,n){switch(e){case"return":t.resolve({value:n,done:!0});break;case"throw":t.reject(n);break;default:t.resolve({value:n,done:!1})}(t=t.next)?o(t.key,t.arg):r=null}this._invoke=function(e,n){return new Promise(function(a,i){var s={key:e,arg:n,resolve:a,reject:i,next:null};r?r=r.next=s:(t=r=s,o(e,n))})},"function"!=typeof e.return&&(this.return=void 0)}r.prototype["function"==typeof Symbol&&Symbol.asyncIterator||"@@asyncIterator"]=function(){return this},r.prototype.next=function(e){return this._invoke("next",e)},r.prototype.throw=function(e){return this._invoke("throw",e)},r.prototype.return=function(e){return this._invoke("return",e)},t.exports=function(e){return function(){return new r(e.apply(this,arguments))}},t.exports.__esModule=!0,t.exports.default=t.exports}}),Oe=I(Me(),1),Ne=I(Be(),1),Ie=I($e(),1);function Pe(e){const t=e[Symbol.asyncIterator]();return t[Symbol.asyncDispose]?t:ke(t,async()=>{var e;await(null===(e=t.return)||void 0===e?void 0:e.call(t))})}function qe(){return(qe=(0,Ie.default)(function*(e,t){try{var n=(0,Oe.default)();const r=n.a(Pe(e)),o=n.u(Ee(t.maxDurationMs)).start();let a;for(;;){if(a=yield(0,Ne.default)(_e.race([r.next(),o])),a===Ce&&de(),a.done)return a;yield a.value,a=null}}catch(e){n.e=e}finally{yield(0,Ne.default)(n.d())}})).apply(this,arguments)}function Ae(){return(Ae=(0,Ie.default)(function*(e,t){try{var n=(0,Oe.default)();const r=n.a(Pe(e));let o;const a=n.u(Ee(t.gracePeriodMs));let i=t.count,s=new Promise(()=>{});for(;;){if(o=yield(0,Ne.default)(_e.race([r.next(),s])),o===Ce&&de(),o.done)return o.value;yield o.value,0===--i&&(s=a.start()),o=null}}catch(e){n.e=e}finally{yield(0,Ne.default)(n.d())}})).apply(this,arguments)}function ze(){let e,t;return{promise:new Promise((n,r)=>{e=n,t=r}),resolve:e,reject:t}}var Fe=I(Me(),1),Ue=I(Be(),1),je=I($e(),1);function De(){let e="idle",t=ze();const n=[],r=new Set,o=[];function a(n){if("pending"!==e)return;const a=function(e,t){const n=e[Symbol.asyncIterator]();let r="idle";function o(){r="done",t=()=>{}}return{pull:function(){"idle"===r&&(r="pending",n.next().then(e=>{if(e.done)return r="done",t({status:"return",value:e.value}),void o();r="idle",t({status:"yield",value:e.value})}).catch(e=>{t({status:"error",error:e}),o()}))},destroy:async()=>{var e;o(),await(null===(e=n.return)||void 0===e?void 0:e.call(n))}}}(n,n=>{if("pending"===e){switch(n.status){case"yield":o.push([a,n]);break;case"return":r.delete(a);break;case"error":o.push([a,n]),r.delete(a)}t.resolve()}});r.add(a),a.pull()}return{add(t){switch(e){case"idle":n.push(t);break;case"pending":a(t)}},[Symbol.asyncIterator]:()=>(0,je.default)(function*(){try{var i=(0,Fe.default)();if("idle"!==e)throw new Error("Cannot iterate twice");e="pending";i.a(ke({},async()=>{e="done";const n=[];if(await Promise.all(Array.from(r.values()).map(async e=>{try{await e.destroy()}catch(e){n.push(e)}})),o.length=0,r.clear(),t.resolve(),n.length>0)throw new AggregateError(n)}));for(;n.length>0;)a(n.shift());for(;r.size>0;){for(yield(0,Ue.default)(t.promise);o.length>0;){const[e,t]=o.shift();switch(t.status){case"yield":yield t.value,e.pull();break;case"error":throw t.error}}t=ze()}}catch(e){i.e=e}finally{yield(0,Ue.default)(i.d())}})()}}function Re(e){const t=e[Symbol.asyncIterator]();return new ReadableStream({async cancel(){var e;await(null===(e=t.return)||void 0===e?void 0:e.call(t))},async pull(e){const n=await t.next();n.done?e.close():e.enqueue(n.value)}})}var Le=I(Me(),1),Ge=I(Be(),1),He=I($e(),1);const Je=Symbol("ping");function Ve(e,t){return Ke.apply(this,arguments)}function Ke(){return(Ke=(0,He.default)(function*(e,t){try{var n=(0,Le.default)();const o=n.a(Pe(e));let a,i=o.next();for(;;)try{var r=(0,Le.default)();const e=r.u(Ee(t));if(a=yield(0,Ge.default)(_e.race([i,e.start()])),a===Ce){yield Je;continue}if(a.done)return a.value;i=o.next(),yield a.value,a=null}catch(e){r.e=e}finally{r.d()}}catch(e){n.e=e}finally{yield(0,Ge.default)(n.d())}})).apply(this,arguments)}var Ye=N({"../../node_modules/.pnpm/@oxc-project+runtime@0.72.2/node_modules/@oxc-project/runtime/src/helpers/asyncIterator.js"(e,t){function n(e){function t(e){if(Object(e)!==e)return Promise.reject(new TypeError(e+" is not an object."));var t=e.done;return Promise.resolve(e.value).then(function(e){return{value:e,done:t}})}return(n=function(e){this.s=e,this.n=e.next}).prototype={s:null,n:null,next:function(){return t(this.n.apply(this.s,arguments))},return:function(e){var n=this.s.return;return void 0===n?Promise.resolve({value:e,done:!0}):t(n.apply(this.s,arguments))},throw:function(e){var n=this.s.return;return void 0===n?Promise.reject(e):t(n.apply(this.s,arguments))}},new n(e)}t.exports=function(e){var t,r,o,a=2;for("undefined"!=typeof Symbol&&(r=Symbol.asyncIterator,o=Symbol.iterator);a--;){if(r&&null!=(t=e[r]))return t.call(e);if(o&&null!=(t=e[o]))return new n(t.call(e));r="@@asyncIterator",o="@@iterator"}throw new TypeError("Object is not async iterable")},t.exports.__esModule=!0,t.exports.default=t.exports}}),We=I(Be(),1),Qe=I($e(),1),Ze=I(Me(),1),Xe=I(Ye(),1);function et(e){return(_(e)||"function"==typeof e)&&"function"==typeof(null==e?void 0:e.then)&&"function"==typeof(null==e?void 0:e.catch)}var tt=class extends Error{constructor(e){super("Max depth reached at path: "+e.join(".")),this.path=e}};function nt(){return nt=(0,Qe.default)(function*(e){const{data:t}=e;let n=0;const r=De();function o(e){const t=n++,o=e(t);return r.add(o),t}function a(t,n){return o(function(){var r=(0,Qe.default)(function*(r){const o=s(n);o&&(t.catch(t=>{var r;null===(r=e.onError)||void 0===r||r.call(e,{error:t,path:n})}),t=Promise.reject(o));try{const e=yield(0,We.default)(t);yield[r,0,c(e,n)]}catch(t){var a,i;null===(a=e.onError)||void 0===a||a.call(e,{error:t,path:n}),yield[r,1,null===(i=e.formatError)||void 0===i?void 0:i.call(e,{error:t,path:n})]}});return function(e){return r.apply(this,arguments)}}())}function i(t,n){return o(function(){var r=(0,Qe.default)(function*(r){try{var o=(0,Ze.default)();const l=s(n);if(l)throw l;const u=o.a(Pe(t));try{for(;;){const e=yield(0,We.default)(u.next());if(e.done){yield[r,0,c(e.value,n)];break}yield[r,1,c(e.value,n)]}}catch(t){var a,i;null===(a=e.onError)||void 0===a||a.call(e,{error:t,path:n}),yield[r,2,null===(i=e.formatError)||void 0===i?void 0:i.call(e,{error:t,path:n})]}}catch(e){o.e=e}finally{yield(0,We.default)(o.d())}});return function(e){return r.apply(this,arguments)}}())}function s(t){return e.maxDepth&&t.length>e.maxDepth?new tt(t):null}function l(t,n){if(et(t))return[0,a(t,n)];if(S(t)){if(e.maxDepth&&n.length>=e.maxDepth)throw new Error("Max depth reached");return[1,i(t,n)]}return null}function c(e,t){if(void 0===e)return[[]];const n=l(e,t);if(n)return[[0],[null,...n]];if(!1===pe(r=e)||void 0!==(o=r.constructor)&&(!1===pe(a=o.prototype)||!1===a.hasOwnProperty("isPrototypeOf")))return[[e]];var r,o,a;const i={},s=[];for(const[n,r]of Object.entries(e)){const e=l(r,[...t,n]);e?(i[n]=0,s.push([n,...e])):i[n]=r}return[[i],...s]}const u={};for(const[e,n]of Object.entries(t))u[e]=c(n,[e]);yield u;let d=r;e.pingMs&&(d=Ve(r,e.pingMs));var p,m=!1,f=!1;try{for(var h,y=(0,Xe.default)(d);m=!(h=yield(0,We.default)(y.next())).done;m=!1){const e=h.value;yield e}}catch(e){f=!0,p=e}finally{try{m&&null!=y.return&&(yield(0,We.default)(y.return()))}finally{if(f)throw p}}}),nt.apply(this,arguments)}function rt(e){let t=Re(function(e){return nt.apply(this,arguments)}(e));const{serialize:n}=e;return n&&(t=t.pipeThrough(new TransformStream({transform(e,t){e===Je?t.enqueue(Je):t.enqueue(n(e))}}))),t.pipeThrough(new TransformStream({transform(e,t){e===Je?t.enqueue(" "):t.enqueue(JSON.stringify(e)+"\n")}})).pipeThrough(new TextEncoderStream)}Error;var ot=N({"../../node_modules/.pnpm/@oxc-project+runtime@0.72.2/node_modules/@oxc-project/runtime/src/helpers/asyncGeneratorDelegate.js"(e,t){var n=Te();t.exports=function(e){var t={},r=!1;function o(t,o){return r=!0,o=new Promise(function(n){n(e[t](o))}),{done:!1,value:new n(o,1)}}return t["undefined"!=typeof Symbol&&Symbol.iterator||"@@iterator"]=function(){return this},t.next=function(e){return r?(r=!1,e):o("next",e)},"function"==typeof e.throw&&(t.throw=function(e){if(r)throw r=!1,e;return o("throw",e)}),"function"==typeof e.return&&(t.return=function(e){return r?(r=!1,e):o("return",e)}),t},t.exports.__esModule=!0,t.exports.default=t.exports}}),at=I(Ye(),1),it=I(Be(),1),st=I($e(),1),lt=I(ot(),1);I(Me(),1);const ct="ping",ut="serialized-error",dt="connected",pt="return";function mt(e){var t,n,r,o,a;const{serialize:i=C}=e,s={enabled:null!==(t=null===(n=e.ping)||void 0===n?void 0:n.enabled)&&void 0!==t&&t,intervalMs:null!==(r=null===(o=e.ping)||void 0===o?void 0:o.intervalMs)&&void 0!==r?r:1e3},l=null!==(a=e.client)&&void 0!==a?a:{};if(s.enabled&&l.reconnectAfterInactivityMs&&s.intervalMs>l.reconnectAfterInactivityMs)throw new Error(`Ping interval must be less than client reconnect interval to prevent unnecessary reconnection - ping.intervalMs: ${s.intervalMs} client.reconnectAfterInactivityMs: ${l.reconnectAfterInactivityMs}`);function c(){return c=(0,st.default)(function*(){yield{event:dt,data:JSON.stringify(l)};let t,n,r=e.data;e.emitAndEndImmediately&&(r=function(e,t){return Ae.apply(this,arguments)}(r,{count:1,gracePeriodMs:1})),e.maxDurationMs&&e.maxDurationMs>0&&e.maxDurationMs!==1/0&&(r=function(e,t){return qe.apply(this,arguments)}(r,{maxDurationMs:e.maxDurationMs})),s.enabled&&s.intervalMs!==1/0&&s.intervalMs>0&&(r=Ve(r,s.intervalMs));var o,a=!1,c=!1;try{for(var u,d=(0,at.default)(r);a=!(u=yield(0,it.default)(d.next())).done;a=!1)t=u.value,t!==Je?(n=X(t)?{id:t[0],data:t[1]}:{data:t},n.data=JSON.stringify(i(n.data)),yield n,t=null,n=null):yield{event:ct,data:""}}catch(e){c=!0,o=e}finally{try{a&&null!=d.return&&(yield(0,it.default)(d.return()))}finally{if(c)throw o}}}),c.apply(this,arguments)}function u(){return u=(0,st.default)(function*(){try{yield*(0,lt.default)((0,at.default)(function(){return c.apply(this,arguments)}())),yield{event:pt,data:""}}catch(r){var t,n;if(ue(r))return;const o=J(r),a=null!==(t=null===(n=e.formatError)||void 0===n?void 0:n.call(e,{error:o}))&&void 0!==t?t:null;yield{event:ut,data:JSON.stringify(i(a))}}}),u.apply(this,arguments)}const d=Re(function(){return u.apply(this,arguments)}());return d.pipeThrough(new TransformStream({transform(e,t){"event"in e&&t.enqueue(`event: ${e.event}\n`),"data"in e&&t.enqueue(`data: ${e.data}\n`),"id"in e&&t.enqueue(`id: ${e.id}\n`),"comment"in e&&t.enqueue(`: ${e.comment}\n`),t.enqueue("\n\n")}})).pipeThrough(new TextEncoderStream)}const ft={"Content-Type":"text/event-stream","Cache-Control":"no-cache, no-transform","X-Accel-Buffering":"no",Connection:"keep-alive"};var ht=I($e(),1),yt=I(D(),1);function gt(e){return k((0,ht.default)(function*(){throw e}))}const bt={mutation:["POST"],query:["GET"],subscription:["GET"]},vt={mutation:["POST"],query:["GET","POST"],subscription:["GET","POST"]};function wt(e){var t,n,r;const{ctx:o,info:a,responseMeta:i,untransformedJSON:s,errors:l=[],headers:c}=e;let u=s?function(e){const t=Array.isArray(e)?e:[e],n=new Set(t.map(e=>{var t;return"error"in e&&_(e.error.data)?"number"==typeof(null===(t=e.error.data)||void 0===t?void 0:t.httpStatus)?e.error.data.httpStatus:q(w[e.error.code]):200}));return 1!==n.size?207:n.values().next().value}(s):200;const d=!s,p=d?[]:Array.isArray(s)?s:[s],m=null!==(t=null==i?void 0:i({ctx:o,info:a,paths:null==a?void 0:a.calls.map(e=>e.path),data:p,errors:l,eagerGeneration:d,type:null!==(n=null==a||null===(r=a.calls.find(e=>{var t;return null===(t=e.procedure)||void 0===t?void 0:t._def.type}))||void 0===r||null===(r=r.procedure)||void 0===r?void 0:r._def.type)&&void 0!==n?n:"unknown"}))&&void 0!==t?t:{};if(m.headers)if(m.headers instanceof Headers)for(const[e,t]of m.headers.entries())c.append(e,t);else for(const[e,t]of Object.entries(m.headers))if(Array.isArray(t))for(const n of t)c.append(e,n);else"string"==typeof t&&c.set(e,t);return m.status&&(u=m.status),{status:u}}function _t(e){return!!_(e)&&(!!S(e)||(Object.values(e).some(et)||Object.values(e).some(S)))}async function xt(e){var t,n,r,o,a,i;const{router:s,req:l}=e,c=new Headers([["vary","trpc-accept"]]),u=s._def._config,d=new URL(l.url);if("HEAD"===l.method)return new Response(null,{status:204});const p=null===(t=null!==(n=e.allowBatching)&&void 0!==n?n:null===(r=e.batching)||void 0===r?void 0:r.enabled)||void 0===t||t,m=null!==(o=e.allowMethodOverride)&&void 0!==o&&o&&"POST"===l.method,f=await k(async()=>{try{return[void 0,await ce({req:l,path:decodeURIComponent(e.path),router:s,searchParams:d.searchParams,headers:e.req.headers,url:d})]}catch(e){return[J(e),void 0]}}),h=k(()=>{let t;return{valueOrUndefined:()=>{if(t)return t[1]},value:()=>{const[e,n]=t;if(e)throw e;return n},create:async n=>{if(t)throw new Error("This should only be called once - report a bug in tRPC");try{const r=await e.createContext({info:n});t=[void 0,r]}catch(e){t=[J(e),void 0]}}}}),y=m?vt:bt,g="application/jsonl"===l.headers.get("trpc-accept"),b=null===(a=null===(i=u.sse)||void 0===i?void 0:i.enabled)||void 0===a||a;try{const[t,n]=f;if(t)throw t;if(n.isBatchCall&&!p)throw new V({code:"BAD_REQUEST",message:"Batching is not enabled on the server"});if(g&&!n.isBatchCall)throw new V({message:"Streaming requests must be batched (you can do a batch of 1)",code:"BAD_REQUEST"});await h.create(n);const r=n.calls.map(async t=>{const r=t.procedure;try{if(e.error)throw e.error;if(!r)throw new V({code:"NOT_FOUND",message:`No procedure found on path "${t.path}"`});if(!y[r._def.type].includes(l.method))throw new V({code:"METHOD_NOT_SUPPORTED",message:`Unsupported ${l.method}-request to ${r._def.type} procedure at path "${t.path}"`});if("subscription"===r._def.type&&n.isBatchCall)throw new V({code:"BAD_REQUEST",message:"Cannot batch subscription calls"});return[void 0,{data:await r({path:t.path,getRawInput:t.getRawInput,ctx:h.value(),type:r._def.type,signal:e.req.signal})}]}catch(n){var o,a,i;const r=J(n),s=t.result();return null===(o=e.onError)||void 0===o||o.call(e,{error:r,path:t.path,input:s,ctx:h.valueOrUndefined(),type:null!==(a=null===(i=t.procedure)||void 0===i?void 0:i._def.type)&&void 0!==a?a:"unknown",req:e.req}),[r,void 0]}});if(!n.isBatchCall){const[t]=n.calls,[o,a]=await r[0];switch(n.type){case"unknown":case"mutation":case"query":{if(c.set("content-type","application/json"),_t(null==a?void 0:a.data))throw new V({code:"UNSUPPORTED_MEDIA_TYPE",message:"Cannot use stream-like response in non-streaming request - use httpBatchStreamLink"});const r=o?{error:L({config:u,ctx:h.valueOrUndefined(),error:o,input:t.result(),path:t.path,type:n.type})}:{result:{data:a.data}},i=wt({ctx:h.valueOrUndefined(),info:n,responseMeta:e.responseMeta,errors:o?[o]:[],headers:c,untransformedJSON:[r]});return new Response(JSON.stringify(W(u,r)),{status:i.status,headers:c})}case"subscription":{const r=k(()=>{if(o)return gt(o);if(!b)return gt(new V({code:"METHOD_NOT_SUPPORTED",message:'Missing experimental flag "sseSubscriptions"'}));if(!ee(a.data)&&!S(a.data))return gt(new V({message:`Subscription ${t.path} did not return an observable or a AsyncGenerator`,code:"INTERNAL_SERVER_ERROR"}));return ee(a.data)?re(a.data,e.req.signal):a.data}),i=mt((0,yt.default)((0,yt.default)({},u.sse),{},{data:r,serialize:e=>u.transformer.output.serialize(e),formatError(n){var r,o,a;const i=J(n.error),s=null==t?void 0:t.result(),l=null==t?void 0:t.path,c=null!==(r=null==t||null===(o=t.procedure)||void 0===o?void 0:o._def.type)&&void 0!==r?r:"unknown";null===(a=e.onError)||void 0===a||a.call(e,{error:i,path:l,input:s,ctx:h.valueOrUndefined(),req:e.req,type:c});return L({config:u,ctx:h.valueOrUndefined(),error:i,input:s,path:l,type:c})}}));for(const[e,t]of Object.entries(ft))c.set(e,t);const s=wt({ctx:h.valueOrUndefined(),info:n,responseMeta:e.responseMeta,errors:[],headers:c,untransformedJSON:null});return new Response(i,{headers:c,status:s.status})}}}if("application/jsonl"===n.accept){c.set("content-type","application/json"),c.set("transfer-encoding","chunked");const t=wt({ctx:h.valueOrUndefined(),info:n,responseMeta:e.responseMeta,errors:[],headers:c,untransformedJSON:null}),o=rt((0,yt.default)((0,yt.default)({},u.jsonl),{},{maxDepth:1/0,data:r.map(async t=>{const[r,o]=await t,a=n.calls[0];var i,s;if(r)return{error:L({config:u,ctx:h.valueOrUndefined(),error:r,input:a.result(),path:a.path,type:null!==(i=null===(s=a.procedure)||void 0===s?void 0:s._def.type)&&void 0!==i?i:"unknown"})};const l=ee(o.data)?re(o.data,e.req.signal):Promise.resolve(o.data);return{result:Promise.resolve({data:l})}}),serialize:e=>u.transformer.output.serialize(e),onError:t=>{var r,o;null===(r=e.onError)||void 0===r||r.call(e,{error:J(t),path:void 0,input:void 0,ctx:h.valueOrUndefined(),req:e.req,type:null!==(o=null==n?void 0:n.type)&&void 0!==o?o:"unknown"})},formatError(e){var t,r;const o=null==n?void 0:n.calls[e.path[0]],a=J(e.error),i=null==o?void 0:o.result(),s=null==o?void 0:o.path,l=null!==(t=null==o||null===(r=o.procedure)||void 0===r?void 0:r._def.type)&&void 0!==t?t:"unknown";return L({config:u,ctx:h.valueOrUndefined(),error:a,input:i,path:s,type:l})}}));return new Response(o,{headers:c,status:t.status})}c.set("content-type","application/json");const o=(await Promise.all(r)).map(e=>{const[t,n]=e;return t?e:_t(n.data)?[new V({code:"UNSUPPORTED_MEDIA_TYPE",message:"Cannot use stream-like response in non-streaming request - use httpBatchStreamLink"}),void 0]:e}),a=o.map(([e,t],r)=>{const o=n.calls[r];var a,i;return e?{error:L({config:u,ctx:h.valueOrUndefined(),error:e,input:o.result(),path:o.path,type:null!==(a=null===(i=o.procedure)||void 0===i?void 0:i._def.type)&&void 0!==a?a:"unknown"})}:{result:{data:t.data}}}),i=o.map(([e])=>e).filter(Boolean),s=wt({ctx:h.valueOrUndefined(),info:n,responseMeta:e.responseMeta,untransformedJSON:a,errors:i,headers:c});return new Response(JSON.stringify(W(u,a)),{status:s.status,headers:c})}catch(t){var v;const[n,r]=f,o=h.valueOrUndefined(),{error:a,untransformedJSON:i,body:s}=function(e,t){const{router:n,req:r,onError:o}=t.opts,a=J(e);null==o||o({error:a,path:t.path,input:t.input,ctx:t.ctx,type:t.type,req:r});const i={error:L({config:n._def._config,error:a,type:t.type,path:t.path,input:t.input,ctx:t.ctx})},s=W(n._def._config,i);return{error:a,untransformedJSON:i,body:JSON.stringify(s)}}(t,{opts:e,ctx:h.valueOrUndefined(),type:null!==(v=null==r?void 0:r.type)&&void 0!==v?v:"unknown"}),l=wt({ctx:o,info:r,responseMeta:e.responseMeta,untransformedJSON:i,errors:[a],headers:c});return new Response(s,{status:l.status,headers:c})}}function St(e){try{var t,n;const r=e.headers[":scheme"]&&"https"===e.headers[":scheme"]||e.socket&&"encrypted"in e.socket&&e.socket.encrypted?"https:":"http:",o=null!==(t=null!==(n=e.headers.host)&&void 0!==n?n:e.headers[":authority"])&&void 0!==t?t:"localhost";return new URL(e.url,`${r}//${o}`)}catch(e){throw new V({code:"BAD_REQUEST",message:"Invalid URL",cause:e})}}function kt(e){const t=new Headers;for(const n in e){const r=e[n];if("string"!=typeof n||!n.startsWith(":"))if(Array.isArray(r))for(const e of r)t.append(n,e);else null!=r&&t.append(n,r)}return t}function Ct(e,t,n){const r=new AbortController,o=()=>{t.off("close",o),e.off("aborted",o),r.abort()};t.once("close",o),e.once("aborted",o);const a=St(e),i={headers:kt(e.headers),method:e.method,signal:r.signal};"GET"!==e.method&&"HEAD"!==e.method&&(i.body=function(e,t){if("body"in e){if(void 0===e.body)return;return"string"==typeof e.body?e.body:JSON.stringify(e.body)}let n=0,r=!1;return new ReadableStream({start(o){const a=s=>{n+=s.length,!t.maxBodySize||n<=t.maxBodySize?o.enqueue(new Uint8Array(s.buffer,s.byteOffset,s.byteLength)):(o.error(new V({code:"PAYLOAD_TOO_LARGE"})),r=!0,e.off("data",a),e.off("end",i))},i=()=>{r||(r=!0,e.off("data",a),e.off("end",i),o.close())};e.on("data",a),e.on("end",i)},cancel(){e.destroy()}})}(e,n),i.duplex="half");return new Request(a,i)}async function Et(e){const{res:t}=e;try{const n=new WritableStream({async write(e){var n;await async function(e,t){!1===e.write(t)&&await new Promise((t,n)=>{const r=e=>{n(e),a()},o=()=>{t(),a()},a=()=>{e.off("error",r),e.off("drain",o)};e.once("error",r),e.once("drain",o)})}(t,e),null===(n=t.flush)||void 0===n||n.call(t)}});await e.body.pipeTo(n,{signal:e.signal})}catch(e){if(ue(e))return;throw e}}var Mt=I(D(),1);function Tt(e){return t=>{var n;const{res:r,req:o}=e,a=J(t),i=L({config:e.router._def._config,error:a,type:"unknown",path:void 0,input:void 0,ctx:void 0});null===(n=e.onError)||void 0===n||n.call(e,{req:o,error:a,type:"unknown",path:void 0,input:void 0,ctx:void 0});const s=W(e.router._def._config,{error:i});r.statusCode=i.data.httpStatus,r.end(JSON.stringify(s))}}async function Bt(e){return new Promise(t=>{var n;const r=null!==(n=e.middleware)&&void 0!==n?n:(e,t,n)=>n();return e.res.once("finish",()=>{t()}),r(e.req,e.res,t=>{k(async()=>{var n;const r=Ct(e.req,e.res,{maxBodySize:null!==(n=e.maxBodySize)&&void 0!==n?n:null}),o=await xt((0,Mt.default)((0,Mt.default)({},e),{},{req:r,error:t?J(t):null,createContext:async t=>{var n;return await(null===(n=e.createContext)||void 0===n?void 0:n.call(e,(0,Mt.default)((0,Mt.default)({},e),t)))},onError(t){var n;null==e||null===(n=e.onError)||void 0===n||n.call(e,(0,Mt.default)((0,Mt.default)({},t),{},{req:e.req}))}}));await async function(e){const{response:t,rawResponse:n}=e;200===n.statusCode&&(n.statusCode=t.status);for(const[e,r]of t.headers)n.setHeader(e,r);try{t.body&&await Et({res:n,signal:e.request.signal,body:t.body})}catch(e){throw n.headersSent||(n.statusCode=500),e}finally{n.end()}}({request:r,response:o,rawResponse:e.res})}).catch(Tt(e))})})}const $t=t("http");var Ot=I(D(),1);function Nt(e){var t;const n=(null!==(t=e.basePath)&&void 0!==t?t:"/").length;return(t,r)=>{let o="";k(async()=>{const a=St(t);o=a.pathname.slice(n),await Bt((0,Ot.default)((0,Ot.default)({},e),{},{req:t,res:r,path:o}))}).catch(Tt((0,Ot.default)({req:t,res:r,path:o},e)))}}function It(e){return $t.createServer(function(e){return Nt(e)}(e))}const Pt=t("node:os"),qt=t("node:crypto"),At=1024**3,zt=(e,t,n)=>Math.min(Math.max(e,t),n),Ft=e=>e?40:0,Ut=(e=0)=>{if(!e)return 0;return zt(e/(12*At)*20,0,20)},jt=(e=0)=>{if(!e)return 0;return zt(e/(32*At)*10,0,10)},Dt=e=>e?10:0,Rt=(e="default",t=null)=>{const n=String(e).toLowerCase();if(!n)return 0;if(n.includes("cuda"))return 20;if(n.includes("vulkan"))return 10;if(n.includes("default")){return"darwin"===t||"ios"===t?15:5}return 0},Lt=({platform:e,variant:t,hasGpu:n,gpuUsableBytes:r=0,cpuUsableBytes:o=0,ok:a=!0}={})=>{if(!a)return 0;const i=Ft(n)+Rt(t,e)+Ut(r),s=jt(o),l=Dt(a);return Math.min(100,Math.round(i+s+l))},Gt=["cuda","vulkan","snapdragon","default"],Ht=e=>e||0===e?Array.isArray(e)?e.filter(e=>null!=e):[e]:[],Jt=e=>e&&String(e).trim().toLowerCase()||null,Vt=(e={})=>{const t=String(e.type||e.deviceType||e.kind||"").toLowerCase();return!!t.includes("gpu")||(!!t.includes("cuda")||(!!t.includes("metal")||(!!t.includes("vulkan")||!!t.includes("snapdragon"))))},Kt=({platform:e,totalMemoryInBytes:t,variant:n,devices:r,gpuMemoryFraction:o,cpuMemoryFraction:a,ok:i,error:s})=>{const l=(e=>Array.isArray(e)?e.map(e=>({...e})):[])(((e,t)=>"snapdragon"===e?t.filter(e=>"GPUOpenCL"!==e.deviceName):t)(n,r)),c=l.some(Vt),u=l.filter(e=>Vt(e)&&Number.isFinite(Number(e.maxMemorySize))).reduce((e,t)=>e+t.maxMemorySize,0),d=t,p=c?Math.floor(u*o):0,m=d?Math.floor(d*a):0,f={platform:e,variant:n,hasGpu:c,gpuUsableBytes:p,cpuUsableBytes:m,ok:i},h=Lt(f),y=i?(({platform:e,variant:t,hasGpu:n,gpuUsableBytes:r=0,cpuUsableBytes:o=0,ok:a=!0}={})=>({gpuPresence:Ft(n),variant:Rt(t,e),gpuMemory:Ut(r),cpuMemory:jt(o),availability:Dt(a)}))(f):null;return{platform:e,ok:i,variant:n,hasGpu:c,devices:l,gpuTotalBytes:u,gpuUsableBytes:p,cpuTotalBytes:d,cpuUsableBytes:m,score:h,breakdown:y,error:s,timestamp:(new Date).toISOString()}},Yt=({device:e,modelBytes:t=0,kvCacheBytes:n=0}={})=>{if(!e)return{totalRequiredBytes:t+n,fitsInGpu:!1,fitsInCpu:!1,limiting:"unknown-device"};const r=Math.max(0,Number(t)||0)+Math.max(0,Number(n)||0),o=e.hasGpu&&r>0&&r<=e.gpuUsableBytes,a=r>0&&r<=e.cpuUsableBytes;let i="ok";return!o&&e.hasGpu&&(i="gpu-memory"),a||(i=o?"cpu-memory":"insufficient-memory"),{totalRequiredBytes:r,fitsInGpu:o,fitsInCpu:a,limiting:i}},Wt=async({platform:e,variant:t=null,preferVariants:n=[],variantPreference:r=[],gpuMemoryFraction:o=.85,cpuMemoryFraction:a=.5,includeBreakdown:i=!1,totalMemoryInBytes:s,modelBytes:l=null,kvCacheBytes:c=null,limitedKvCacheBytes:u=null,dependencies:d={},defaultVariants:p=Gt}={})=>{const{getBackendDevicesInfo:m,isLibVariantAvailable:f}=d;if("function"!=typeof m||"function"!=typeof f)throw new TypeError("GGML capability detection requires getBackendDevicesInfo and isLibVariantAvailable functions");const h=(({variant:e,preferVariants:t=[],variantPreference:n=[],defaultVariants:r=Gt}={})=>{const o=[];e&&o.push(e),o.push(...Ht(t)),o.push(...Ht(n)),o.push(...r);const a=o.map(Jt).filter(Boolean);return Array.from(new Set(a))})({variant:t,preferVariants:n,variantPreference:r,defaultVariants:p}),y=[];for(const t of h)try{if(!await f(t))throw new Error(`Variant ${t} not available on this platform`);const n=await m(t);y.push(Kt({platform:e,totalMemoryInBytes:s,variant:t,devices:n,gpuMemoryFraction:o,cpuMemoryFraction:a,ok:!0}))}catch(n){const r=n instanceof Error?n.message:String(n);y.push(Kt({platform:e,totalMemoryInBytes:s,variant:t,devices:[],gpuMemoryFraction:o,cpuMemoryFraction:a,ok:!1,error:r}))}const g=y.filter(e=>e.ok)[0]||null,b={ok:Boolean(g),selected:g?{...g,breakdown:i?g.breakdown:void 0}:null,attempts:y};if(!i&&b.selected&&delete b.selected.breakdown,!b||!l&&!c)return b;const v=e=>{if(!e)return e;const t=Yt({device:e,modelBytes:l||0,kvCacheBytes:c||0});let n=null;return null!=u&&u!==c&&(n=Yt({device:e,modelBytes:l||0,kvCacheBytes:u})),{...e,fit:t,...n&&{limitedFit:n}}};return b.selected=v(b.selected),b.attempts=Array.isArray(b.attempts)?b.attempts.map(v):b.attempts,b},Qt="ggml-llm",Zt=["cuda","vulkan","default"],Xt=new Map([[Qt,Wt],["ggml-stt",async({platform:e,variant:t=null,preferVariants:n=[],variantPreference:r=[],gpuMemoryFraction:o=.85,cpuMemoryFraction:a=.5,includeBreakdown:i=!1,totalMemoryInBytes:s,modelBytes:l=null,processingBytes:c=null,kvCacheBytes:u=null,dependencies:d={}}={})=>{const p=r&&r.length>0?r:Zt;return Wt({platform:e,variant:t,preferVariants:n,variantPreference:p,gpuMemoryFraction:o,cpuMemoryFraction:a,includeBreakdown:i,totalMemoryInBytes:s,modelBytes:l,kvCacheBytes:c??u,dependencies:d,defaultVariants:Zt})}]]),en=async({platform:e,totalMemoryInBytes:t,backend:n=Qt,dependencies:r,...o}={})=>{const a=Xt.get(n);if(!a)throw new Error(`No capability detector registered for backend "${n}"`);return await a({...o,dependencies:r,totalMemoryInBytes:t,platform:e})},tn={f16:2,f32:4,q8_0:1,q6_k:.75,q5_k:.625,q5_k_m:.625,q5_k_s:.625,q5_1:.625,q5_0:.625,q4_k:.5,q4_k_m:.5,q4_k_s:.5,q4_1:.5,q4_0:.5,iq4_nl:.5},nn=e=>{const t=e?String(e).toLowerCase():"f16";return tn[t]||tn.f16},rn=({modelBytes:e=0,audioLengthSeconds:t=30,sampleRate:n=16e3,bytesPerSample:r=4}={})=>{const o=Math.max(0,Number(e)||0),a=Math.max(0,Math.floor(Math.max(0,t)*n*r)),i=1048576,s=1024*i;let l,c,u;l=o<200*i?120*i:o<500*i?140*i:o<2*s?150*i:160*i,c=o<200*i?70*i:o<500*i?135*i:220*i,u=o<100*i?20*i:o<200*i?30*i:o<500*i?85*i:o<2*s?215*i:360*i;const d=l+c+u;return{modelBytes:o,audioBufferBytes:a,processingBufferBytes:d,totalBytes:o+d+a}},on=e=>e?String(e).trim().toLowerCase():null,an=({arch:e,nLayer:t=0})=>({arch:on(e),enabled:!1,window:null,pattern:null,denseFirst:!1,type:null,kvLayers:Math.max(0,Math.floor(Number(t)||0)),swaLayers:0}),sn=new Map([["llama4",({nSwa:e})=>{if(0===e)return{enabled:!1};return{enabled:!0,window:e&&e>0?e:8192,pattern:4,type:"chunked"}}],["afmoe",({nSwa:e})=>!e||e<=0?{enabled:!1}:{enabled:!0,window:e,pattern:4,type:"standard"}],["phi3",()=>({enabled:!1})],["gemma2",({nSwa:e})=>{const t=e&&e>0?e:4096;return t?{enabled:!0,window:t,pattern:2,type:"standard"}:{enabled:!1}}],["gemma3",({nSwa:e})=>!e||e<=0?{enabled:!1}:{enabled:!0,window:e,pattern:6,type:"standard"}],["gemma3n",({nLayer:e,nSwa:t})=>!t||t<=0?{enabled:!1}:{enabled:!0,window:t,pattern:5,type:"standard",kvLayers:Math.min(20,e)}],["gemma-embedding",({nSwa:e})=>!e||e<=0?{enabled:!1}:{enabled:!0,window:e,pattern:6,type:"symmetric"}],["cohere2",({nSwa:e})=>!e||e<=0?{enabled:!1}:{enabled:!0,window:e,pattern:4,type:"standard"}],["olmo2",({nSwa:e})=>!e||e<=0?{enabled:!1}:{enabled:!0,window:e,pattern:4,type:"standard"}],["exaone4",({nLayer:e,nSwa:t})=>{let n=null;return t&&t>0?n=t:e>=64&&(n=4096),n?{enabled:!0,window:n,pattern:4,type:"standard"}:{enabled:!1}}],["gpt-oss",({nSwa:e})=>!e||e<=0?{enabled:!1}:{enabled:!0,window:e,pattern:2,type:"standard"}],["smallthinker",({nSwa:e})=>!e||e<=0?{enabled:!1}:{enabled:!0,window:4096,pattern:4,denseFirst:!0,type:"standard"}]]),ln=({arch:e,metadata:t={},nLayer:n=0}={})=>{const r=on(e||t["general.architecture"]),o=Math.max(0,Math.floor(Number(n)||0)),a=((e={},t=null)=>{if(!e)return null;const n=on(t),r=n?`${n}.attention.sliding_window`:null,o=(r&&null!=e[r]?e[r]:null)??e["llama.attention.sliding_window"];if(null==o)return null;const a=Number(o);return Number.isFinite(a)?a:null})(t,r),i=r?sn.get(r):null;if(!i)return an({arch:r,nLayer:n});const s=i({nLayer:o,nSwa:a,metadata:t});if(!s||!s.enabled||!s.window||s.window<=0)return an({arch:r,nLayer:n});const l=Math.max(0,Math.floor(Number(s.pattern)||0)),c=null!=s.kvLayers&&Number.isFinite(Number(s.kvLayers))?Number(s.kvLayers):o,u=Math.max(0,Math.floor(c)),d=((e=0,t=0,n=!1)=>{const r=Math.max(0,Math.floor(Number(e)||0)),o=Math.max(0,Math.floor(Number(t)||0));if(!r||1===o)return 0;if(o<=0)return r;const a=Math.max(0,o-1),i=r%o;return Math.floor(r/o)*a+(n?Math.max(0,i-1):Math.min(i,a))})(u,l,Boolean(s.denseFirst));return{arch:r,enabled:d>0,window:s.window,pattern:l,denseFirst:Boolean(s.denseFirst),type:s.type||"standard",kvLayers:u,swaLayers:d}},cn=(e={})=>{const t=e["general.architecture"],n=(t,n=null)=>{const r=e[t],o=Number(r);return Number.isFinite(o)?o:n},r=t?n(`${t}.context_length`,n("llama.context_length")):null,o=t?n(`${t}.block_count`,n("llama.block_count")):null,a=t?n(`${t}.embedding_length`,n("llama.embedding_length")):null,i=t?n(`${t}.attention.head_count`,n("llama.attention.head_count")):null;return{arch:t,nCtxTrain:r,nLayer:o,nEmbd:a,nHead:i,nHeadKv:t?n(`${t}.attention.head_count_kv`,n("llama.attention.head_count_kv",i)):null,nEmbdHeadK:t?n(`${t}.attention.key_length`,n("llama.attention.key_length")):null,nEmbdHeadV:t?n(`${t}.attention.value_length`,n("llama.attention.value_length")):null,quantVersion:e["general.quantization_version"]||null,fileType:e["general.file_type"]||null}},un=({layerCount:e,headKvCount:t,embdHeadKCount:n,embdHeadVCount:r,cacheTypes:o,swaConfig:a,kvUnified:i=!1,nParallel:s=1,swaFull:l=!1})=>{const c=a?.window&&i?Math.max(1,Number(s)||1):1;return i=>((e,t,n,r,o,a={},{totalLayers:i=null,swaLayers:s=0,swaContext:l=null,swaContextMultiplier:c=1,swaAdditionalTokens:u=0,swaFull:d=!1}={})=>{if(!(e&&t&&n&&r&&o))return 0;const p=Number(null!=i&&void 0!==i?i:e),m=Math.max(0,Math.floor(p));if(!m)return 0;const f=nn(a.k),h=nn(a.v),y=Number(n)*(Number(r)*f+Number(o)*h);if(!y)return 0;const g=Math.max(0,Number(t)||0),b=Math.min(m,Math.max(0,Math.floor(Number(s)||0))),v=Math.max(0,m-b),w=(null!=l&&Number.isFinite(Number(l))?Math.max(0,Number(l)):g)*Math.max(1,Number(c)||1)+Math.max(0,Number(u)||0),_=d?g:Math.min(g,w),x=v*g+b*Math.max(0,Math.floor(_));return Math.round(y*x)})(e,i,t,n,r,o,{totalLayers:e,swaLayers:a?.swaLayers||0,swaContext:a?.window,swaFull:l,swaContextMultiplier:c})},dn=({maxCtx:e,availableMemory:t,modelBytes:n,kvBytesForCtx:r})=>{const o=Math.max(1,Math.floor(Number(e)||0));if(!r||t<=n)return o;let a=1,i=o,s=o;for(;a<=i;){const e=Math.floor((a+i)/2);n+r(e)<=t?(s=e,a=e+1):i=e-1}return s};var pn=f(285);const mn=t("node:stream/web");var fn=f.t(mn,2);const hn=t("node:events"),yn=new hn.EventEmitter;yn.setMaxListeners(100);const gn=(e,t,n)=>{e.push({...t,timestamp:t.timestamp||(new Date).toISOString()}),e.length>n&&e.shift()};class bn{constructor(e=9999){this.maxEntries=e,this.modelLoads=[],this.completions=[],this.transcriptions=[]}addModelLoad(e){gn(this.modelLoads,e,this.maxEntries),yn.emit("status:modelLoad",e),yn.emit("status:change",{type:"modelLoad",entry:e})}addCompletion(e){gn(this.completions,e,this.maxEntries),yn.emit("status:completion",e),yn.emit("status:change",{type:"completion",entry:e})}addTranscription(e){gn(this.transcriptions,e,this.maxEntries),yn.emit("status:transcription",e),yn.emit("status:change",{type:"transcription",entry:e})}getModelLoadHistory(){return[...this.modelLoads].reverse()}getCompletionHistory(){return[...this.completions].reverse()}getTranscriptionHistory(){return[...this.transcriptions].reverse()}clear(){this.modelLoads=[],this.completions=[],this.transcriptions=[]}}const vn=new bn,wn=new bn;let _n=0;function xn(e){const t=t=>e(t);return yn.on("status:change",t),()=>yn.off("status:change",t)}function Sn(e){const t=[];return{generators:Array.from(e.entries()).filter(([,e])=>"ggml-llm"===e.type).map(([e,n])=>{const{instance:r}=n;let o=[];return r.contexts&&(o=Array.from(r.contexts.entries()).map(([n,r])=>{const o={key:n,refCount:r.refCount,hasModel:Boolean(r.context)},a=r.context.parallel.getStatus();return o.parallelStatus=a,t.push({generatorId:e,contextKey:n,...a}),o})),{id:e,type:n.type,refCount:n.refCount,repoId:r.info?.model?.repoId||null,quantization:r.info?.model?.quantization||null,variant:r.info?.runtime?.variant||null,nCtx:r.info?.runtime?.n_ctx||null,nParallel:r.info?.runtime?.n_parallel||null,contexts:o}}),parallelStatuses:t,history:{modelLoads:vn.getModelLoadHistory(),completions:vn.getCompletionHistory()}}}function kn(e){return{generators:Array.from(e.entries()).filter(([,e])=>"ggml-stt"===e.type).map(([e,t])=>{const{instance:n}=t,r=n.getStatus?.()||{},o=r.queueStatus||{processing:!1,queuedCount:0};return{id:e,type:t.type,refCount:t.refCount,repoId:n.info?.model?.repoId||null,quantization:n.info?.model?.quantization||null,variant:n.info?.runtime?.variant||null,hasContext:r.hasContext||!1,contextRefCount:r.contextRefCount||0,queueStatus:o}}),history:{modelLoads:wn.getModelLoadHistory(),transcriptions:wn.getTranscriptionHistory()}}}const Cn="undefined"!=typeof globalThis&&globalThis.ReadableStream&&globalThis.WritableStream?{ReadableStream:globalThis.ReadableStream,WritableStream:globalThis.WritableStream}:fn,{ReadableStream:En,WritableStream:Mn}=Cn,Tn=new TextEncoder,Bn=(e={},t={})=>(Object.entries(t||{}).forEach(([t,n])=>{n&&"object"==typeof n&&!Array.isArray(n)?(e[t]&&"object"==typeof e[t]||(e[t]={}),Bn(e[t],n)):e[t]=n}),e),$n="https://huggingface.co",On="https://huggingface.co/api",Nn=g.join(Pt.homedir(),".buttress","models"),In=["mxfp4","q8_0","q6_k","q6","q5_k_m","q5_k_s","q5_k","q5_1","q5_0","q4_k_m","q4_k_s","q4_k","q4_1","q4_0","q3","q2"],Pn=.5,qn={backend:{type:"ggml-llm",variant:null,variant_preference:["cuda","vulkan","snapdragon","default"],gpu_memory_fraction:.85,cpu_memory_fraction:Pn},model:{repo_id:null,revision:"main",filename:null,url:null,quantization:null,preferred_quantizations:[],n_ctx:null,n_gpu_layers:"auto",n_parallel:4,n_batch:2048,flash_attn:"auto",cache_type_k:"f16",cache_type_v:"f16",ctx_shift:!1,kv_unified:!1,swa_full:!1,allow_local_file:!1,local_path:null,api_base:On,base_url:$n},runtime:{cache_dir:Nn,prefer_variants:[],huggingface_token:process.env.HUGGINGFACE_TOKEN||null,http_headers:{},session_cache:{enabled:!0,max_size_bytes:10737418240,max_entries:1e3}}},An=(e,t=[])=>e||0===e?Array.isArray(e)?e.filter(e=>null!=e):[e]:[...t],zn=e=>{if(!e)return null;const t=String(e).toLowerCase();return["cuda","vulkan","snapdragon","default"].includes(t)?t:null},Fn=(e={})=>{const t=JSON.parse(JSON.stringify(qn));if(Bn(t,e),t.backend.variant=zn(t.backend.variant),t.backend.variant_preference=Array.from(new Set(An(t.backend.variant_preference).map(zn).filter(Boolean))),0===t.backend.variant_preference.length&&(t.backend.variant_preference=["cuda","vulkan","snapdragon","default"]),t.runtime.prefer_variants=Array.from(new Set(An(t.runtime.prefer_variants).map(zn).filter(Boolean))),t.model.preferred_quantizations=Array.from(new Set(An(t.model.preferred_quantizations||t.model.quantizations).map(e=>e?String(e).toLowerCase():null).filter(Boolean))),t.model.quantization){const e=String(t.model.quantization).toLowerCase();t.model.preferred_quantizations.includes(e)||t.model.preferred_quantizations.unshift(e)}return t.model.n_parallel=Math.max(1,Number(t.model.n_parallel)||4),t.model.n_batch=Math.max(1,Number(t.model.n_batch)||512),t.model.base_url=t.model.base_url||$n,t.model.api_base=t.model.api_base||On,t.runtime.cache_dir=t.runtime.cache_dir?g.resolve(t.runtime.cache_dir):Nn,t.runtime.session_cache={...qn.runtime.session_cache,...t.runtime.session_cache||{}},t},Un=e=>{const t=e.toLowerCase();return In.find(e=>t.includes(e))||null},jn=async e=>{await(0,y.mkdir)(e,{recursive:!0})},Dn=(e,t,n=Nn)=>{const r=(0,qt.createHash)("sha256").update(e).digest("hex");return g.join(((e=Nn)=>g.join(e,".metadata-cache"))(n),t,`${r}.json`)},Rn=async(e,t,n=Nn)=>{try{const r=Dn(e,t,n),o=await(0,y.readFile)(r,"utf-8");return console.log(`[Cache] Hit ${t} cache:`,g.basename(r)),JSON.parse(o,(e,t)=>"string"==typeof t&&t.startsWith("__bigint__")?BigInt(t.slice(10)):t)}catch(e){return null}},Ln=async(e,t,n,r=Nn)=>{try{const o=Dn(e,t,r);await jn(g.dirname(o)),await(0,y.writeFile)(o,JSON.stringify(n,(e,t)=>"bigint"==typeof t?`__bigint__${t.toString()}`:t),"utf-8"),console.log(`[Cache] Wrote ${t} cache:`,g.basename(o))}catch(e){console.warn(`[Cache] Failed to write ${t} cache:`,e.message)}},Gn=(e=Nn)=>g.join(e,".session-state-cache"),Hn=(e=Nn)=>g.join(Gn(e),"cache-map.json"),Jn=(e=Nn)=>g.join(Gn(e),"temp"),Vn=(e=Nn)=>g.join(Gn(e),"states"),Kn=async(e,t=Nn)=>{const n=Hn(t),r=`${n}.tmp.${Date.now()}`;try{await jn(g.dirname(n)),await(0,y.writeFile)(r,JSON.stringify(e,null,2),"utf-8"),await(0,y.rename)(r,n)}catch(e){throw await(0,y.unlink)(r).catch(()=>{}),e}},Yn=(e,t,n)=>{const r=Object.values(n.entries);console.log(`[SessionCache] Finding match for promptText (${e.length} chars)`),console.log(`[SessionCache] Checking ${r.length} cache entries`);const o=r.filter(e=>{return n=e.metadata,r=t,n.modelPath===r.modelPath&&n.variant===r.variant&&n.n_gpu_layers===r.n_gpu_layers&&n.n_ctx>=r.n_ctx&&n.cacheTypeK===r.cacheTypeK&&n.cacheTypeV===r.cacheTypeV;var n,r}).reduce((t,n)=>{const r=((e,t)=>{const n=Math.min(e.length,t.length);let r=0;for(;r<n&&e[r]===t[r];)r+=1;return r})(e,n.fullText);return r>=100&&r>t.prefixLen?{entry:n,prefixLen:r}:t},{entry:null,prefixLen:0});return o.entry?(console.log(`[SessionCache] Prefix match found: ${o.entry.id} (${o.prefixLen}/${o.entry.fullText.length} chars)`),{entry:o.entry,prefixLength:o.prefixLen}):(console.log("[SessionCache] No match found"),null)};class Wn{constructor(e,t){this.config=e,this.plan=t,this.baseDir=e.runtime.cache_dir,this.enabled=!1!==e.runtime.session_cache?.enabled,this.maxSizeBytes=((e,t)=>{if(null==e)return t;if("number"==typeof e)return e;if("string"==typeof e){const n=pn.parse(e);return null!=n?n:t}return t})(e.runtime.session_cache?.max_size_bytes,10737418240),this.maxEntries=e.runtime.session_cache?.max_entries||1e3,this.metadata={variant:t.info?.runtime?.variant||null,n_gpu_layers:t.info?.runtime?.n_gpu_layers||0,n_ctx:t.info?.runtime?.n_ctx||0,modelPath:t.localPath,cacheTypeK:t.info?.runtime?.cache_type_k||"f16",cacheTypeV:t.info?.runtime?.cache_type_v||"f16"},this.cacheMap=null,this.initialized=!1}async initialize(){if(this.enabled&&!this.initialized)try{await jn(Gn(this.baseDir)),await jn(Jn(this.baseDir)),await jn(Vn(this.baseDir)),this.cacheMap=await(async(e=Nn)=>{try{const t=Hn(e),n=await(0,y.readFile)(t,"utf-8"),r=JSON.parse(n);return r.entries&&"object"==typeof r.entries?r:{version:1,entries:{},totalSize:0}}catch{return{version:1,entries:{},totalSize:0}}})(this.baseDir),this.initialized=!0,console.log(`[SessionCache] Initialized with ${Object.keys(this.cacheMap.entries).length} entries`)}catch(e){console.warn(`[SessionCache] Failed to initialize: ${e.message}`),this.enabled=!1}}async findMatchingEntry(e){if(!this.enabled||!this.cacheMap)return null;const t=Yn(e,this.metadata,this.cacheMap);if(t){const{entry:e}=t;return await(async e=>{try{return await(0,y.stat)(e),!0}catch{return!1}})(e.stateFilePath)?(e.lastAccessedAt=(new Date).toISOString(),await Kn(this.cacheMap,this.baseDir).catch(()=>{}),{entry:e}):(console.log(`[SessionCache] Removing stale entry: ${e.id}`),delete this.cacheMap.entries[e.id],this.cacheMap.totalSize-=e.stateFileSize||0,await Kn(this.cacheMap,this.baseDir).catch(()=>{}),null)}return null}async prepareCompletionOptions(e,t){if(!this.enabled)return{options:e,cacheEntry:null,promptPrefix:null};const n=await this.findMatchingEntry(t);if(n){const{entry:t}=n;console.log(`[SessionCache] Found matching entry: ${t.id} (${t.fullText.length} chars, loadStateSize=${t.loadStateSize})`);return{options:{...e,load_state_path:t.stateFilePath},cacheEntry:t,promptPrefix:t.fullText}}return{options:e,cacheEntry:null,promptPrefix:null}}async saveCompletionState(e,t,n,r=0){if(!this.enabled)return null;const o=((e,t)=>{const n=JSON.stringify({text:e,model:t.modelPath,variant:t.variant,n_gpu_layers:t.n_gpu_layers,n_ctx:t.n_ctx});return(0,qt.createHash)("sha256").update(n).digest("hex").slice(0,24)})(e,this.metadata);if(this.cacheMap.entries[o])return console.log(`[SessionCache] Entry already exists for prompt: ${o}`),await(0,y.unlink)(n).catch(()=>{}),this.cacheMap.entries[o];const a=e+t,i=((e,t=Nn)=>g.join(Vn(t),`${e}.bin`))(o,this.baseDir);try{await jn(g.dirname(i)),await(0,y.rename)(n,i);const s=await(0,y.stat)(i),l={id:o,promptText:e,completionText:t,fullText:a,promptTokenCount:r,stateFilePath:i,stateFileSize:s.size,metadata:{...this.metadata},createdAt:(new Date).toISOString(),lastAccessedAt:(new Date).toISOString()};return this.cacheMap.entries[o]=l,this.cacheMap.totalSize+=s.size,await(async(e,t,n)=>{const r=Object.values(e.entries).sort((e,t)=>new Date(e.lastAccessedAt)-new Date(t.lastAccessedAt));let o=e.totalSize,a=Object.keys(e.entries).length;const i=r.filter(e=>(o>t||a>n)&&(o-=e.stateFileSize||0,a-=1,!0));return await Promise.all(i.map(async t=>{await(0,y.unlink)(t.stateFilePath).catch(()=>{}),delete e.entries[t.id],console.log(`[SessionCache] Evicted entry: ${t.id}`)})),e.totalSize=Math.max(0,o),i.map(e=>e.id)})(this.cacheMap,this.maxSizeBytes,this.maxEntries),await Kn(this.cacheMap,this.baseDir),console.log(`[SessionCache] Saved entry: ${o} (${s.size} bytes)`),l}catch(e){return console.warn(`[SessionCache] Failed to save state: ${e.message}`),await(0,y.unlink)(n).catch(()=>{}),null}}async generateTempStatePath(){return await jn(Jn(this.baseDir)),((e=Nn)=>{const t=`${Date.now()}-${Math.random().toString(36).slice(2,10)}`;return g.join(Jn(e),`${t}.bin`)})(this.baseDir)}async cleanup(){await(async(e=Nn)=>{const t=Jn(e);try{const e=await(0,y.readdir)(t),n=Date.now(),r=36e5;await Promise.all(e.map(async e=>{const o=g.join(t,e),a=await(0,y.stat)(o).catch(()=>null);a&&n-a.mtimeMs>r&&(await(0,y.unlink)(o).catch(()=>{}),console.log(`[SessionCache] Cleaned up temp file: ${e}`))}))}catch{}})(this.baseDir)}}const Qn=async(e,t={})=>{if("function"!=typeof fetch)throw new Error("Global fetch is not available in this runtime");const n=await fetch(e,t);if(!n.ok){const t=await n.text().catch(()=>"");throw new Error(`Failed to fetch ${e}: ${n.status} ${n.statusText} ${t}`.trim())}return n.json()},Zn=async(e,t={})=>{if("function"!=typeof fetch)throw new Error("Global fetch is not available in this runtime");const n=await fetch(e,{...t,method:"HEAD"});if(!n.ok)throw new Error(`Failed to fetch headers for ${e}: ${n.status} ${n.statusText}`);return n},Xn=async(e,t,n=Nn)=>{const r=JSON.stringify({url:e,headers:t}),a=await Rn(r,"range-metadata",n);if(a)return a;const i=!/^https?:/i.test(e),{metadata:s}=await o(e,{fetch,additionalFetchHeaders:t,allowLocalFile:i});return await Ln(r,"range-metadata",s,n),s},er=async(e,t)=>{try{const n=await(0,y.stat)(e);return!t||n.size===t}catch(e){return!1}},tr=async(e,t,n,r,o)=>{if("function"!=typeof fetch)throw new Error("Global fetch is not available in this runtime");await jn(g.dirname(n));const a=await fetch(e,{headers:t});if(!a.ok||!a.body)throw new Error(`Failed to download ${e}: ${a.status} ${a.statusText}`);const i=await(0,y.open)(n,"w"),s=Number(a.headers.get("content-length"))||r||0;let l=0,c=.05;try{await a.body.pipeTo(new Mn({async write(e){if(await i.write(e),l+=e.byteLength,"function"==typeof o&&s>0){const e=Math.min(1,l/s);for(;e>=c;)o(c),c+=.05}},async close(){await i.close(),"function"==typeof o&&o(1)},async abort(e){throw await i.close().catch(()=>{}),await(0,y.unlink)(n).catch(()=>{}),e}}))}catch(e){throw await i.close().catch(()=>{}),await(0,y.unlink)(n).catch(()=>{}),e}if(r){const e=await(0,y.stat)(n);if(e.size!==r)throw await(0,y.unlink)(n).catch(()=>{}),new Error(`Downloaded file size mismatch, expected ${r} got ${e.size}`)}},nr=async e=>{const t=e.model.repo_id||e.model.repository||e.model.model;if(!t)throw new Error("`model.repo_id` is required in Buttress backend config");const n=e.model.revision||"main",r=e.runtime.cache_dir,o=JSON.stringify({repoId:t,revision:n,filename:e.model.filename,url:e.model.url,quantization:e.model.quantization,preferred_quantizations:e.model.preferred_quantizations}),a=await Rn(o,"artifact-info",r);if(a)return a;const i={...e.runtime.http_headers||{}};if(e.runtime.huggingface_token&&(i.Authorization=`Bearer ${e.runtime.huggingface_token}`),e.model.url){const a=await Zn(e.model.url,{headers:i}),s=Number(a.headers.get("content-length"))||null,l={repoId:t,revision:n,filename:e.model.filename||e.model.url.split("/").pop(),url:e.model.url,size:s,headers:i};return await Ln(o,"artifact-info",l,r),l}let{filename:s}=e.model,l=e.model.quantization&&String(e.model.quantization).toLowerCase();const c=await Qn(`${e.model.api_base}/models/${t}?revision=${n}&blobs=true`,{headers:i}),u=(c?.siblings||c?.files||[]).map(e=>e.rfilename||e.path||e.filename).filter(e=>"string"==typeof e&&e.endsWith(".gguf"));if(0===u.length)throw new Error(`No GGUF artifacts found in repo ${t}`);const d=e.model.preferred_quantizations.length>0?e.model.preferred_quantizations:In;if(s)l||(l=Un(s));else{const e=(()=>{const e=d.find(e=>u.find(t=>t.toLowerCase().includes(e)));if(e)return{filename:u.find(t=>t.toLowerCase().includes(e)),quantization:e};return null})()||{filename:u[0],quantization:null},{filename:t,quantization:n}=e;s=t,l=n||Un(s)}const p=`${e.model.base_url.replace(/\/+$/,"")}/${t}/resolve/${n}/${s}`,m=/-(\d{5})-of-(\d{5})\.gguf$/,f=s.match(m);let h=null;if(f){const[,,r]=f,o=await Qn(`${e.model.api_base}/models/${t}?revision=${n}&blobs=true`,{headers:i}),a=o?.siblings||o?.files||[],l=Number(r);h=0;for(let e=1;e<=l;e+=1){const t=String(e).padStart(5,"0"),n=s.replace(m,`-${t}-of-${r}.gguf`),o=a.find(e=>(e.rfilename||e.path||e.filename)===n),i=Number(o?.size);Number.isFinite(i)&&i>0&&(h+=i)}}else{const e=await Zn(p,{headers:i});h=Number(e.headers.get("content-length"))||null}const y={repoId:t,revision:n,filename:s,url:p,size:h,quantization:l,headers:i,isSplit:Boolean(f),splitCount:f?Number(f[2]):0};return await Ln(o,"artifact-info",y,r),y},rr=async(e,{modelBytes:t=null,kvCacheBytes:n=null}={})=>{const r=(e=>{const t=[];return e.backend.variant&&t.push(e.backend.variant),e.runtime.prefer_variants.length>0&&t.push(...e.runtime.prefer_variants),t.push(...e.backend.variant_preference),t.push("default"),Array.from(new Set(t.map(zn).filter(Boolean)))})(e),[o,...s]=r,l=null!=e.backend?.gpu_memory_fraction?Math.min(1,Math.max(0,Number(e.backend.gpu_memory_fraction))):qn.backend.gpu_memory_fraction||1,c=null!=e.backend?.cpu_memory_fraction?Math.min(1,Math.max(0,Number(e.backend.cpu_memory_fraction))):Pn,u=await en({platform:process.platform,totalMemoryInBytes:Pt.totalmem(),backend:"ggml-llm",variant:o||null,preferVariants:s,gpuMemoryFraction:l,cpuMemoryFraction:c,dependencies:{getBackendDevicesInfo:a,isLibVariantAvailable:i},modelBytes:t,kvCacheBytes:n}),d=e=>({...e,devices:Array.isArray(e.devices)?e.devices:[],ok:e.ok,hasGpu:Boolean(e.hasGpu),totalMemory:e.gpuTotalBytes||e.totalMemory||0,error:e.ok?null:new Error(e.error||`Variant ${e.variant} not available on this platform`)});if(!u.ok||!u.selected){const e=(u.attempts||[]).map(e=>`${e.variant}: ${e.error||"unknown error"}`).join("; ");throw new Error(`Unable to initialize any backend variant (${r.join(", ")}). Errors: ${e}`)}const p=(u.attempts||[]).map(d);return{selected:d(u.selected),attempts:p}},or=async e=>{const t=await nr(e),n=await Xn(t.url,t.headers,e.runtime.cache_dir),{arch:r,nCtxTrain:o,nLayer:a,nEmbd:i,nHead:s,nHeadKv:l,nEmbdHeadK:c,nEmbdHeadV:u,quantVersion:d,fileType:p}=cn(n),m=Number.isFinite(Number(a))?Number(a):0,f=Number.isFinite(Number(i))?Number(i):0,h=Number.isFinite(Number(s))?Number(s):0,y=Number.isFinite(Number(l))?Number(l):h,b=h>0&&f>0?f/h:128,v=null!=c&&Number.isFinite(Number(c))?Number(c):b,w=null!=u&&Number.isFinite(Number(u))?Number(u):b,_=ln({arch:r,metadata:n,nLayer:m}),x=_&&Number.isFinite(Number(_.kvLayers))?Number(_.kvLayers):m,S=Math.max(0,Math.floor(Number(x)||0)),k=e.model.n_ctx?Number(e.model.n_ctx):null;let C=k||o||4096;const E=[],M=[];let T=!0;if(k&&o&&k>o){T=!1;const e=`Requested context length (${k}) exceeds model training context (${o})`;E.push(e),M.push(e),C=o}k&&!o&&E.push("Model metadata missing training context length, using requested value");const B={k:e.model.cache_type_k,v:e.model.cache_type_v},$=t.size>0?t.size:0,O=un({layerCount:S,headKvCount:y,embdHeadKCount:v,embdHeadVCount:w,cacheTypes:B,swaConfig:_,kvUnified:e.model.kv_unified,nParallel:e.model.n_parallel,swaFull:e.model.swa_full}),N=O(C),I=await rr(e,{modelBytes:$,kvCacheBytes:N}),P=I.selected.totalMemory||0,q=P*(e.backend.gpu_memory_fraction||1),A=null!=e.backend.cpu_memory_fraction?Math.min(1,Math.max(0,Number(e.backend.cpu_memory_fraction))):Pn,z=Math.max(0,Pt.totalmem()*A),F=I.selected.hasGpu?q:z,U=dn({maxCtx:C,availableMemory:F,modelBytes:$,kvBytesForCtx:O});if(!k&&U){const e=o?Math.min(U,o):U,t=Math.max(32,e);t<C&&E.push(`Context length capped to ${t} by memory limits`),C=t}C>U&&(C=U);const j=Math.floor(U);console.log(`[buttress] Memory-limited context length: ${j}`);const D=O(C),R=$+D,L=m?$/(m+1):$;let G,H=0;I.selected.hasGpu&&L>0&&(H=Math.min(m+1,Math.max(0,Math.floor(q/L)))),console.log(`[buttress] Auto GPU layer capacity (${I.selected.variant}): ${H}/${m+1}`),G="auto"===e.model.n_gpu_layers||null==e.model.n_gpu_layers?H:Math.max(0,Math.min(Number(e.model.n_gpu_layers)||0,m+1));const J=(()=>{const t=e.model.flash_attn&&String(e.model.flash_attn).toLowerCase();return"on"===t||"off"===t?t:I.selected.hasGpu?"auto":"off"})(),V=e.runtime.cache_dir,K=((e,t)=>{if(e.model.local_path)return g.resolve(e.model.local_path);const n=t.repoId.split("/"),r=g.join(e.runtime.cache_dir,...n,t.revision);return g.join(r,t.filename)})(e,t),Y=await er(K,t.size),W={ok:T,backend:"ggml-llm",warnings:E,errors:M,model:{repoId:t.repoId,revision:t.revision,filename:t.filename,quantization:t.quantization,url:t.url,sizeBytes:t.size,metadata:{architecture:r,n_ctx_train:o,n_layer:m,n_embd:f,quantization_version:d,file_type:p,kv_layer_count:S,swa:_?.enabled?{window:_.window,pattern:_.pattern,dense_first:_.denseFirst,type:_.type,layers:_.swaLayers}:null}},runtime:{variant:I.selected.variant,n_ctx:C,requested_ctx:k,n_gpu_layers:G,n_parallel:e.model.n_parallel,n_batch:e.model.n_batch,flash_attn_type:J,cache_type_k:B.k,cache_type_v:B.v,ctx_shift:e.model.ctx_shift,kv_unified:e.model.kv_unified,swa_full:e.model.swa_full,estimated_max_n_ctx:j,auto_gpu_layers:H},resources:{modelBytes:$,kvCacheBytes:D,totalEstimatedBytes:R,gpuCapacityBytes:P,gpuUsableBytes:q,cpuUsableBytes:z,fit:I.selected.fit},devices:{selected:I.selected,attempts:I.attempts},download:{cacheDir:V,localPath:K,exists:Y},timestamp:(new Date).toISOString()};return{config:e,info:W,artifact:t,metadata:{arch:r,nCtxTrain:o,nLayer:m,nEmbd:f},devices:I,cacheTypes:B,localPath:K,localExists:Y}},ar=(e,t)=>`event: ${e}\ndata: ${null==t?"":JSON.stringify(t)}\n\n`,ir=(e,t,n=null)=>{let r;const o=Date.now();let a=0;return new En({async start(i){try{const s=await e.parallel.completion(t,(e,t)=>{t&&(t.token&&(a+=1),i.enqueue(Tn.encode(ar("token",{requestId:e,...t}))))}),{requestId:l}=s;r=s.stop;const c=await s.promise;console.log("[Completion] Result:",c),i.enqueue(Tn.encode(ar("result",{requestId:l,...c}))),i.close();const u=Date.now()-o,d=c.timings||{};vn.addCompletion({id:`completion-${l}`,generatorId:n,requestId:l,promptTokens:d.prompt_n??0,tokensGenerated:d.predicted_n??a,tokensPerSecond:d.predicted_per_second??0,promptPerSecond:d.prompt_per_second??0,durationMs:u,success:!0,interrupted:c.interrupted||!1,contextFull:c.context_full||c.contextFull||!1})}catch(e){i.enqueue(Tn.encode(ar("error",{message:e?.message||String(e)}))),i.error(e),vn.addCompletion({id:`completion-${Date.now()}`,generatorId:n,durationMs:Date.now()-o,tokensGenerated:a,success:!1,error:e?.message||String(e)})}},cancel(){r&&r()}})},sr=(e,t,n,r,o,a,i=null)=>{let s,l="",c=!1;const u=Date.now();let d=0;return new En({async start(p){try{const m=await e.parallel.completion(t,(e,t)=>{t&&(t.token&&(l+=t.token,d+=1),p.enqueue(Tn.encode(ar("token",{requestId:e,...t}))))}),{requestId:f}=m;s=m.stop;const h=await m.promise;h.text?l=h.text:h.content&&(l=h.content),c=!h.interrupted&&!h.context_full,console.log("[Completion] Result:",h),p.enqueue(Tn.encode(ar("result",{requestId:f,...h}))),p.close();const g=Date.now()-u,b=h.timings||{};vn.addCompletion({id:`completion-${f}`,generatorId:i,requestId:f,promptTokens:b.prompt_n??a??0,tokensGenerated:b.predicted_n??d,tokensPerSecond:b.predicted_per_second??0,promptPerSecond:b.prompt_per_second??0,durationMs:g,success:!0,interrupted:h.interrupted||!1,contextFull:h.context_full||h.contextFull||!1,usedCache:Boolean(t.load_state_path)}),c&&n.enabled&&l?n.saveCompletionState(r,l,o,a).catch(e=>{console.warn("[SessionCache] Save failed:",e.message)}):o&&(0,y.unlink)(o).catch(()=>{})}catch(e){p.enqueue(Tn.encode(ar("error",{message:e?.message||String(e)}))),p.error(e),vn.addCompletion({id:`completion-${Date.now()}`,generatorId:i,durationMs:Date.now()-u,tokensGenerated:d,success:!1,error:e?.message||String(e)}),(0,y.unlink)(o).catch(()=>{})}},cancel(){s&&s(),(0,y.unlink)(o).catch(()=>{})}})},lr=e=>{const t={model:e.plan.localPath,runtime:e.plan.info.runtime};return(0,qt.createHash)("sha256").update(JSON.stringify(t)).digest("hex").slice(0,24)},cr=async(e,t)=>{const n=lr(e);let r=e.contexts.get(n);if(r)return r.refCount+=1,await r.ready,"function"==typeof t&&t(1),r;r={key:n,refCount:1},r.ready=(async()=>{const o=Date.now(),a=await(async(e,t,n)=>{const{config:r,localPath:o,artifact:a}=e;if(e.localExists&&!t.has(o))return e.info.download.exists=!0,"function"==typeof n&&n(.5),o;if(r.model.local_path&&!r.model.allow_local_file)throw new Error("Local model path provided but `model.allow_local_file` is not enabled");const i=o;t.has(i)||t.set(i,(async()=>{if(a.isSplit&&a.splitCount>0){const e=/-(\d{5})-of-(\d{5})\.gguf$/,t=g.dirname(o),i=a.splitCount;let s=0;for(let o=1;o<=i;o+=1){const l=String(o).padStart(5,"0"),c=a.filename.replace(e,`-${l}-of-${String(i).padStart(5,"0")}.gguf`),u=`${r.model.base_url.replace(/\/+$/,"")}/${a.repoId}/resolve/${a.revision}/${c}`,d=g.join(t,c);await er(d)||await tr(u,a.headers,d,null,e=>{if(e>=0&&Number.isFinite(e)){const t=(s+e)/i,r=Math.round(100*t);console.log(`Downloading model splits: ${Math.min(100,r)}%`),"function"==typeof n&&n(.5*t)}}),s+=1}}else console.log("Downloading model: 0%"),await tr(a.url,a.headers,o,a.size,e=>{if(e>=0&&Number.isFinite(e)){const t=Math.round(100*e);console.log(`Downloading model: ${Math.min(100,t)}%`),"function"==typeof n&&n(.5*e)}});e.localExists=!0,e.info.download.exists=!0})());try{await t.get(i)}finally{t.delete(i)}return o})(e.plan,e.downloads,t);"function"==typeof t&&t(.5);const i={model:a,n_ctx:e.plan.info.runtime.n_ctx,n_gpu_layers:e.plan.info.runtime.n_gpu_layers,n_parallel:e.plan.info.runtime.n_parallel,n_batch:e.plan.info.runtime.n_batch,flash_attn_type:e.plan.info.runtime.flash_attn_type,ctx_shift:e.plan.info.runtime.ctx_shift,kv_unified:e.plan.info.runtime.kv_unified,swa_full:e.plan.info.runtime.swa_full,lib_variant:e.plan.info.runtime.variant};let l;"off"!==e.plan.info.runtime.flash_attn_type&&(i.cache_type_k=e.plan.info.runtime.cache_type_k,i.cache_type_v=e.plan.info.runtime.cache_type_v),console.log("[Context] Load Options:",i);try{if(l=await s(i,e=>{"function"==typeof t&&(t(.5+.25*e),e%5==0&&console.log("[Context] Load Model Progress:",e))}),e.plan.info.runtime.n_parallel){if(!await l.parallel.enable({n_parallel:e.plan.info.runtime.n_parallel,n_batch:e.plan.info.runtime.n_batch}))throw new Error("Failed to enable parallel decoding mode for context")}return"function"==typeof t&&t(1),r.context=l,r.modelInfo=l.getModelInfo(),vn.addModelLoad({id:`${e.id}-${n}`,generatorId:e.id,contextKey:n,repoId:e.plan.info.model?.repoId||null,quantization:e.plan.info.model?.quantization||null,variant:e.plan.info.runtime?.variant||null,nCtx:e.plan.info.runtime?.n_ctx||null,nGpuLayers:e.plan.info.runtime?.n_gpu_layers||null,durationMs:Date.now()-o,success:!0}),r}catch(t){if(vn.addModelLoad({id:`${e.id}-${n}`,generatorId:e.id,contextKey:n,repoId:e.plan.info.model?.repoId||null,quantization:e.plan.info.model?.quantization||null,variant:e.plan.info.runtime?.variant||null,durationMs:Date.now()-o,success:!1,error:t?.message||String(t)}),l)try{l.release()}catch(e){}throw t}})(),e.contexts.set(n,r);try{return await r.ready,r}catch(t){throw e.contexts.delete(n),t}},ur=async(e,t,n=!1)=>{if(n)t.refCount=0;else if(t.refCount=Math.max(0,t.refCount-1),t.refCount>0)return!1;e.contexts.delete(t.key);try{t.context?.parallel?.disable?.()}catch(e){}return await(t.context?.release?.()),!0};async function dr(e,t){const n=Fn(t),r=await or(n),o=new Wn(n,r);await o.initialize();const a={id:e,type:"ggml-llm",config:n,plan:r,info:r.info,contexts:new Map,downloads:new Map,sessionCache:o};return{id:e,type:"ggml-llm",info:r.info,contexts:a.contexts,initContext:async(e={})=>{const{onProgress:t}=e,n=await cr(a,t);return{modelInfo:n.modelInfo?{...n.modelInfo}:null,runtime:{...a.plan.info.runtime},download:{...a.plan.info.download}}},completion:async(e={})=>{const{options:t={},useCache:n=!0}=e,r=lr(a),o=a.contexts.get(r);if(!o)throw new Error(`Context "${r}" not initialized`);await o.ready;let i=t.prompt||"";if(!i&&t.messages){const e=await o.context.getFormattedChat(t.messages,t.chat_template||t.chatTemplate,{jinja:t.jinja??!0,tools:t.tools,parallel_tool_calls:t.parallel_tool_calls,tool_choice:t.tool_choice,enable_thinking:t.enable_thinking,add_generation_prompt:t.add_generation_prompt,now:t.now,chat_template_kwargs:t.chat_template_kwargs});i=e?.prompt||e||""}if(n&&a.sessionCache.enabled&&i){const{options:e}=await a.sessionCache.prepareCompletionOptions(t,i),n=await a.sessionCache.generateTempStatePath(),r=await o.context.tokenize(i),s=r?.tokens?.length||0,l={...e,save_state_path:n,save_state_size:s};return console.log(`[SessionCache] save_state_size=${s} (prompt tokens)`),sr(o.context,l,a.sessionCache,i,n,s,a.id)}return ir(o.context,t,a.id)},tokenize:async(e={})=>{const{text:t="",params:n={}}=e,r=lr(a),o=a.contexts.get(r);if(!o)throw new Error(`Context "${r}" not initialized`);await o.ready;const i=await o.context.tokenize(t,n);if(!i)return{tokens:[]};const s=Array.from(i.tokens??[]).map(e=>Number(e));return{...i,tokens:s}},detokenize:async(e={})=>{const{tokens:t=[]}=e,n=lr(a),r=a.contexts.get(n);if(!r)throw new Error(`Context "${n}" not initialized`);await r.ready;const o=t.map(e=>Number(e));return r.context.detokenize(o)},applyChatTemplate:async(e={})=>{const{messages:t=[],template:n,params:r}=e,o=lr(a),i=a.contexts.get(o);if(!i)throw new Error(`Context "${o}" not initialized`);return await i.ready,await i.context.getFormattedChat(t,n,r)},releaseContext:async(e={})=>{const t=lr(a),n=a.contexts.get(t);return!!n&&ur(a,n,Boolean(e.force))},finalize:async()=>{const e=Array.from(a.contexts.values()).map(e=>ur(a,e,!0));await Promise.allSettled(e),await a.sessionCache.cleanup()},getStatus:()=>{const e=[],t=Array.from(a.contexts.entries()).map(([t,n])=>{const r={key:t,refCount:n.refCount,hasModel:Boolean(n.context)},o=n.context.parallel.getStatus();return r.parallelStatus=o,e.push({contextKey:t,...o}),r});return{id:a.id,type:a.type,repoId:a.plan.info.model?.repoId||null,quantization:a.plan.info.model?.quantization||null,variant:a.plan.info.runtime?.variant||null,nCtx:a.plan.info.runtime?.n_ctx||null,nParallel:a.plan.info.runtime?.n_parallel||null,contexts:t,parallelStatuses:e}},subscribeParallelStatus:e=>{const t=Array.from(a.contexts.entries()).map(([t,n])=>n.context.parallel.subscribeToStatus(n=>{e({contextKey:t,...n})}));return{remove:()=>{t.forEach(e=>{e?.remove&&e.remove()})}}}}}const pr=e=>{const t=Fn(e);return t.model.repo_id||t.model.repository||t.model.model||null};const mr=e=>e?"number"==typeof e.score&&Number.isFinite(e.score)?Number(e.score):Lt(e):0;async function fr(e=null,t={}){const{threshold:n=1.1,includeBreakdown:r=!1,config:o,...s}=t;let l=null,c=null,u=null,d=null,p=null,m=null;if(o)try{const{modelBytes:e,kvCacheBytes:t,limitedKvCacheBytes:n,memoryLimitedCtx:r,kvInfo:a,quantization:i}=await async function(e){const t=Fn(e),n=await nr(t),r=await Xn(n.url,n.headers,t.runtime.cache_dir),{arch:o,nCtxTrain:a,nLayer:i,nEmbd:s,nHead:l,nHeadKv:c,nEmbdHeadK:u,nEmbdHeadV:d,quantVersion:p,fileType:m}=cn(r),f=Number.isFinite(Number(i))?Number(i):0,h=Number.isFinite(Number(s))?Number(s):0,y=Number.isFinite(Number(l))?Number(l):0,g=Number.isFinite(Number(c))?Number(c):y,b=y>0&&h>0?h/y:128,v=null!=u&&Number.isFinite(Number(u))?Number(u):b,w=null!=d&&Number.isFinite(Number(d))?Number(d):b,_=ln({arch:o,metadata:r,nLayer:f}),x=_&&Number.isFinite(Number(_.kvLayers))?Number(_.kvLayers):f,S=Math.max(0,Math.floor(Number(x)||0)),k=(t.model.n_ctx?Number(t.model.n_ctx):null)||a||4096,C={k:t.model.cache_type_k,v:t.model.cache_type_v},E=n.size>0?n.size:0,M=un({layerCount:S,headKvCount:g,embdHeadKCount:v,embdHeadVCount:w,cacheTypes:C,swaConfig:_,kvUnified:t.model.kv_unified,nParallel:t.model.n_parallel,swaFull:t.model.swa_full}),T=null!=t.backend?.gpu_memory_fraction?Math.min(1,Math.max(0,Number(t.backend.gpu_memory_fraction))):qn.backend.gpu_memory_fraction||1,B=null!=t.backend?.cpu_memory_fraction?Math.min(1,Math.max(0,Number(t.backend.cpu_memory_fraction))):Pn,$=M(k),O=await rr(t,{modelBytes:E,kvCacheBytes:$}),N=(O.selected.totalMemory||0)*T,I=Math.max(0,Pt.totalmem()*B),P=O.selected.hasGpu?N:I,q=dn({maxCtx:k,availableMemory:P,modelBytes:E,kvBytesForCtx:M}),A=M(k),z=M(q);return{kvInfo:{nCtxTrain:a,nLayer:f,nEmbd:h,nHeadKv:g,nEmbdHeadK:v,nEmbdHeadV:w,nHeadCount:y,nHeadKvCount:g,kvLayerCount:S,swa:_?.enabled?{window:_.window,pattern:_.pattern,denseFirst:_.denseFirst,type:_.type,layers:_.swaLayers}:null},modelBytes:E,kvCacheBytes:A,limitedKvCacheBytes:z,memoryLimitedCtx:q,quantization:{name:n.quantization||null,fileType:m,version:p}}}(o);l=e,c=t,u=n,d=r,p=a,m=i}catch(e){}const f=null!=o?.backend?.gpu_memory_fraction?Math.min(1,Math.max(0,Number(o.backend.gpu_memory_fraction))):void 0,h=null!=o?.backend?.cpu_memory_fraction?Math.min(1,Math.max(0,Number(o.backend.cpu_memory_fraction))):void 0,y=await en({...s,platform:process.platform,totalMemoryInBytes:Pt.totalmem(),backend:"ggml-llm",includeBreakdown:r,gpuMemoryFraction:f,cpuMemoryFraction:h,dependencies:{getBackendDevicesInfo:a,isLibVariantAvailable:i},modelBytes:l,kvCacheBytes:c,limitedKvCacheBytes:u}),g=y.selected,b=mr(g);g.modelBytes=l||null,g.kvCacheBytes=c||null,g.memoryLimitedCtx=d||null,g.limitedKvCacheBytes=u||null,g.kvInfo=p||null,g.quantization=m||null;let v=null,w=null;if(e){const t=mr(e);w={...e,score:t};let r="buttress",o="buttress-higher-score";if(y.ok)if(t||0===t){const e=w.fit,a=w.limitedFit,i=g?.fit,s=g?.limitedFit,l=e?.fitsInGpu||e?.fitsInCpu||a?.fitsInGpu||a?.fitsInCpu,c=i?.fitsInGpu||i?.fitsInCpu||s?.fitsInGpu||s?.fitsInCpu;l&&!c?(r="local",o="client-fits-in-memory"):c&&!l?(r="buttress",o="buttress-fits-in-memory"):t>b*n?(r="local",o="client-better"):b>t*n?(r="buttress",o="buttress-better"):(r="either",o="comparable-scores")}else r="buttress",o="missing-client-score";else r="local",o="buttress-unavailable";v={buttressScore:b,clientScore:t,threshold:n,recommendation:r,reason:o}}y.ok||v||(v={buttressScore:b,clientScore:e?.score??null,threshold:n,recommendation:"local",reason:"buttress-unavailable"});let _=null;return o&&(_={repoId:o.model?.repo_id||null,quantization:o.model?.quantization||null,nCtx:o.model?.n_ctx||null,cacheKType:o.model?.cache_type_k||"f16",cacheVType:o.model?.cache_type_v||"f16"}),{type:"ggml-llm",timestamp:(new Date).toISOString(),buttress:y,client:w,comparison:v,modelConfig:_}}const{WritableStream:hr}="undefined"!=typeof globalThis&&globalThis.ReadableStream&&globalThis.WritableStream?{ReadableStream:globalThis.ReadableStream,WritableStream:globalThis.WritableStream}:fn,yr=(e={},t={})=>(Object.entries(t||{}).forEach(([t,n])=>{n&&"object"==typeof n&&!Array.isArray(n)?(e[t]&&"object"==typeof e[t]||(e[t]={}),yr(e[t],n)):e[t]=n}),e),gr="https://huggingface.co",br="https://huggingface.co/api",vr=g.join(Pt.homedir(),".buttress","models"),wr=["cuda","vulkan","default"],_r=["q8_0","q5_1","q5_0","q4_1","q4_0","base"],xr={backend:{type:"ggml-stt",variant:null,variant_preference:wr,gpu_memory_fraction:.85,cpu_memory_fraction:.5},model:{repo_id:"BricksDisplay/whisper-ggml",revision:"main",filename:null,url:null,quantization:null,preferred_quantizations:["q5_1","q8_0","base"],allow_local_file:!1,local_path:null,api_base:br,base_url:gr,use_gpu:!0,use_flash_attn:"auto"},runtime:{cache_dir:vr,prefer_variants:[],huggingface_token:process.env.HUGGINGFACE_TOKEN||null,http_headers:{},max_threads:null}},Sr=(e,t=[])=>e||0===e?Array.isArray(e)?e.filter(e=>null!=e):[e]:[...t],kr=e=>{if(!e)return null;const t=String(e).toLowerCase();return["cuda","vulkan","default"].includes(t)?t:null},Cr=(e={})=>{const t=JSON.parse(JSON.stringify(xr));if(yr(t,e),t.backend.variant=kr(t.backend.variant),t.backend.variant_preference=Array.from(new Set(Sr(t.backend.variant_preference||wr).map(kr).filter(Boolean))),0===t.backend.variant_preference.length&&(t.backend.variant_preference=[...wr]),t.runtime.prefer_variants=Array.from(new Set(Sr(t.runtime.prefer_variants).map(kr).filter(Boolean))),t.model.preferred_quantizations=Array.from(new Set(Sr(t.model.preferred_quantizations||t.model.quantizations).map(e=>e?String(e).toLowerCase():null).filter(Boolean))),t.model.quantization){const e=String(t.model.quantization).toLowerCase();t.model.preferred_quantizations.includes(e)||t.model.preferred_quantizations.unshift(e)}return t.model.base_url=t.model.base_url||gr,t.model.api_base=t.model.api_base||br,t.runtime.cache_dir=t.runtime.cache_dir?g.resolve(t.runtime.cache_dir):vr,t},Er=e=>{const t=e.toLowerCase();return _r.find(e=>t.includes(e))||null},Mr=async e=>{await(0,y.mkdir)(e,{recursive:!0})},Tr=(e,t,n=vr)=>{const r=(0,qt.createHash)("sha256").update(e).digest("hex");return g.join(((e=vr)=>g.join(e,".metadata-cache"))(n),t,`${r}.json`)},Br=async(e,t,n,r=vr)=>{try{const o=Tr(e,t,r);await Mr(g.dirname(o)),await(0,y.writeFile)(o,JSON.stringify(n),"utf-8")}catch(e){}},$r=async(e,t={})=>{if("function"!=typeof fetch)throw new Error("Global fetch is not available in this runtime");const n=await fetch(e,{...t,method:"HEAD"});if(!n.ok)throw new Error(`Failed to fetch headers for ${e}: ${n.status} ${n.statusText}`);return n},Or=async(e,t)=>{try{const n=await(0,y.stat)(e);return!t||n.size===t}catch(e){return!1}},Nr=async e=>{const t=e.model.repo_id||e.model.repository||e.model.model;if(!t)throw new Error("`model.repo_id` is required in Buttress backend config");const n=e.model.revision||"main",r=e.runtime.cache_dir,o=JSON.stringify({repoId:t,revision:n,filename:e.model.filename,url:e.model.url,quantization:e.model.quantization,preferred_quantizations:e.model.preferred_quantizations}),a=await(async(e,t,n=vr)=>{try{const r=Tr(e,t,n),o=await(0,y.readFile)(r,"utf-8");return JSON.parse(o)}catch(e){return null}})(o,"artifact-info",r);if(a)return a;const i={...e.runtime.http_headers||{}};if(e.runtime.huggingface_token&&(i.Authorization=`Bearer ${e.runtime.huggingface_token}`),e.model.url){const a=await $r(e.model.url,{headers:i}),s=Number(a.headers.get("content-length"))||null,l=e.model.filename||e.model.url.split("/").pop(),c={repoId:t,revision:n,filename:l,url:e.model.url,size:s,quantization:Er(l||""),headers:i};return await Br(o,"artifact-info",c,r),c}let{filename:s}=e.model,l=e.model.quantization&&String(e.model.quantization).toLowerCase();const c=await(async(e,t={})=>{if("function"!=typeof fetch)throw new Error("Global fetch is not available in this runtime");const n=await fetch(e,t);if(!n.ok){const t=await n.text().catch(()=>"");throw new Error(`Failed to fetch ${e}: ${n.status} ${n.statusText} ${t}`.trim())}return n.json()})(`${e.model.api_base}/models/${t}?revision=${n}&blobs=true`,{headers:i}),u=(c?.siblings||c?.files||[]).map(e=>e.rfilename||e.path||e.filename).filter(e=>"string"==typeof e&&e.endsWith(".bin"));if(0===u.length)throw new Error(`No model artifacts found in repo ${t}`);const d=e.model.preferred_quantizations.length>0?e.model.preferred_quantizations:_r;if(s)l||(l=Er(s));else{const e=(()=>{const e=d.find(e=>u.find(t=>t.toLowerCase().includes(e)));if(e)return{filename:u.find(t=>t.toLowerCase().includes(e)),quantization:e};return null})()||{filename:u[0],quantization:null},{filename:t,quantization:n}=e;s=t,l=n||Er(s)}const p=`${e.model.base_url.replace(/\/+$/,"")}/${t}/resolve/${n}/${s}`,m=await $r(p,{headers:i}),f={repoId:t,revision:n,filename:s,url:p,size:Number(m.headers.get("content-length"))||null,quantization:l,headers:i,isSplit:!1,splitCount:0};return await Br(o,"artifact-info",f,r),f},Ir=async(e,{modelBytes:t=null,processingBytes:n=null}={})=>{const r=(e=>{const t=[];return e.backend.variant&&t.push(e.backend.variant),e.runtime.prefer_variants.length>0&&t.push(...e.runtime.prefer_variants),t.push(...e.backend.variant_preference),t.push("default"),Array.from(new Set(t.map(kr).filter(Boolean)))})(e),[o,...s]=r,l=null!=e.backend?.gpu_memory_fraction?Math.min(1,Math.max(0,Number(e.backend.gpu_memory_fraction))):xr.backend.gpu_memory_fraction||1,c=null!=e.backend?.cpu_memory_fraction?Math.min(1,Math.max(0,Number(e.backend.cpu_memory_fraction))):.5,u=await en({platform:process.platform,totalMemoryInBytes:Pt.totalmem(),backend:"ggml-stt",variant:o||null,preferVariants:s,variantPreference:e.backend.variant_preference,gpuMemoryFraction:l,cpuMemoryFraction:c,dependencies:{getBackendDevicesInfo:a,isLibVariantAvailable:i},modelBytes:t,kvCacheBytes:n}),d=e=>({...e,devices:Array.isArray(e.devices)?e.devices:[],ok:e.ok,hasGpu:Boolean(e.hasGpu),totalMemory:e.gpuTotalBytes||e.totalMemory||0,error:e.ok?null:new Error(e.error||`Variant ${e.variant} not available on this platform`)});if(!u.ok||!u.selected){const e=(u.attempts||[]).map(e=>`${e.variant}: ${e.error||"unknown error"}`).join("; ");throw new Error(`Unable to initialize any backend variant (${r.join(", ")}). Errors: ${e}`)}const p=(u.attempts||[]).map(d);return{selected:d(u.selected),attempts:p}},Pr=async e=>{const t=await Nr(e),n=rn({modelBytes:t.size>0?t.size:0}),r=await Ir(e,{modelBytes:n.modelBytes,processingBytes:n.processingBufferBytes});let o=r.selected.hasGpu&&(void 0===r.selected.fit?.fitsInGpu||r.selected.fit.fitsInGpu);!1===e.model.use_gpu&&(o=!1);const a=e.model.use_flash_attn&&String(e.model.use_flash_attn).toLowerCase();let i;i="on"===a||"true"===a||"off"!==a&&"false"!==a&&o;const s=e.runtime.cache_dir,l=((e,t)=>{if(e.model.local_path)return g.resolve(e.model.local_path);const n=t.repoId.split("/"),r=g.join(e.runtime.cache_dir,...n,t.revision);return g.join(r,t.filename)})(e,t),c=await Or(l,t.size);return{config:e,info:{ok:!0,backend:"ggml-stt",model:{repoId:t.repoId,revision:t.revision,filename:t.filename,quantization:t.quantization,url:t.url,sizeBytes:t.size},runtime:{variant:r.selected.variant,use_gpu:o,use_flash_attn:i,max_threads:e.runtime.max_threads?Number(e.runtime.max_threads):null},resources:{...n,gpuCapacityBytes:r.selected.gpuTotalBytes,gpuUsableBytes:r.selected.gpuUsableBytes,cpuUsableBytes:r.selected.cpuUsableBytes,fit:r.selected.fit},devices:{selected:r.selected,attempts:r.attempts},download:{cacheDir:s,localPath:l,exists:c},timestamp:(new Date).toISOString()},artifact:t,memory:n,devices:r,localPath:l,localExists:c}},qr=async(e,t,n)=>{const{localPath:r,artifact:o,config:a}=e;if(e.localExists)return"function"==typeof n&&n(1),r;const i=t.get(r);if(i)return await i,"function"==typeof n&&n(1),r;const s=(async()=>{if(a.model.allow_local_file){if(!await Or(r,o.size))throw new Error(`Local model file not found: ${r}`);return r}return await(async(e,t,n,r,o)=>{if("function"!=typeof fetch)throw new Error("Global fetch is not available in this runtime");await Mr(g.dirname(n));const a=await fetch(e,{headers:t});if(!a.ok||!a.body)throw new Error(`Failed to download ${e}: ${a.status} ${a.statusText}`);const i=await(0,y.open)(n,"w"),s=Number(a.headers.get("content-length"))||r||0;let l=0,c=.05;try{await a.body.pipeTo(new hr({async write(e){if(await i.write(e),l+=e.byteLength,"function"==typeof o&&s>0){const e=Math.min(1,l/s);for(;e>=c;)o(c),c+=.05}},async close(){await i.close(),"function"==typeof o&&o(1)},async abort(e){throw await i.close().catch(()=>{}),await(0,y.unlink)(n).catch(()=>{}),e}}))}catch(e){throw await i.close().catch(()=>{}),await(0,y.unlink)(n).catch(()=>{}),e}if(r){const e=await(0,y.stat)(n);if(e.size!==r)throw await(0,y.unlink)(n).catch(()=>{}),new Error(`Downloaded file size mismatch, expected ${r} got ${e.size}`)}})(o.url,o.headers,r,o.size,n),r})();t.set(r,s);try{return await s,r}finally{t.delete(r)}};class Ar{constructor(){this.queue=[],this.processing=!1,this.currentTaskId=null}async enqueue(e,t=null){return new Promise((n,r)=>{this.queue.push({task:e,resolve:n,reject:r,taskId:t}),this.processNext()})}async processNext(){if(this.processing||0===this.queue.length)return;this.processing=!0;const{task:e,resolve:t,reject:n,taskId:r}=this.queue.shift();this.currentTaskId=r;try{t(await e())}catch(e){n(e)}finally{this.processing=!1,this.currentTaskId=null,this.processNext()}}getStatus(){return{processing:this.processing,queuedCount:this.queue.length,currentTaskId:this.currentTaskId}}}const zr=e=>{if(!e)return null;if(e instanceof ArrayBuffer)return e;if(ArrayBuffer.isView(e))return e.buffer;if("string"==typeof e){const t=e.startsWith("data:")?e.split(",")[1]||"":e,n=Buffer.from(t,"base64");return n.buffer.slice(n.byteOffset,n.byteOffset+n.byteLength)}throw new Error("Unsupported audioData format, expected base64 string or ArrayBuffer")},Fr=async(e,t)=>{if(e.contextRecord)return e.contextRecord.refCount+=1,await e.contextRecord.ready,"function"==typeof t&&t(1),e.contextRecord;const n={refCount:1};n.ready=(async()=>{const r=Date.now();try{const o=await qr(e.plan,e.downloads,t),a=await l({filePath:o,useFlashAttn:e.plan.info.runtime.use_flash_attn,useGpu:e.plan.info.runtime.use_gpu},e.plan.info.runtime.variant);n.context=a;try{n.modelInfo=await a.getModelInfo()}catch(e){n.modelInfo=null}return wn.addModelLoad({id:e.id,generatorId:e.id,repoId:e.plan.info.model?.repoId||null,quantization:e.plan.info.model?.quantization||null,variant:e.plan.info.runtime?.variant||null,useGpu:e.plan.info.runtime?.use_gpu||!1,durationMs:Date.now()-r,success:!0}),n}catch(t){throw wn.addModelLoad({id:e.id,generatorId:e.id,repoId:e.plan.info.model?.repoId||null,quantization:e.plan.info.model?.quantization||null,variant:e.plan.info.runtime?.variant||null,durationMs:Date.now()-r,success:!1,error:t?.message||String(t)}),t}})(),e.contextRecord=n;try{return await n.ready,"function"==typeof t&&t(1),n}catch(t){throw e.contextRecord=null,t}},Ur=async(e,t,n=!1)=>{if(n)t.refCount=0;else if(t.refCount=Math.max(0,t.refCount-1),t.refCount>0)return!1;return e.contextRecord=null,await(t.context?.release?.()),!0};async function jr(e,t){const n=await Pr(Cr(t)),r={id:e,type:"ggml-stt",config:t,plan:n,info:n.info,contextRecord:null,downloads:new Map,queue:new Ar},o=async(e={})=>{const{audioPath:t,audioData:n,options:o={}}=e,a=await Fr(r),i={...o};r.plan.info.runtime.max_threads&&null==i.maxThreads&&(i.maxThreads=r.plan.info.runtime.max_threads);const s=`transcription-${Date.now()}-${Math.random().toString(36).slice(2,8)}`,l=Date.now();return r.queue.enqueue(async()=>{await a.ready;try{let e;if(n){const t=zr(n),{promise:r}=a.context.transcribeData(t,i);e=await r}else{if(!t)throw new Error("audioPath or audioData is required for transcription");const n=g.resolve(t),{promise:r}=a.context.transcribe(n,i);e=await r}return wn.addTranscription({id:s,generatorId:r.id,durationMs:Date.now()-l,segmentCount:e?.segments?.length||0,textLength:e?.text?.length||0,success:!0}),e}catch(e){throw wn.addTranscription({id:s,generatorId:r.id,durationMs:Date.now()-l,success:!1,error:e?.message||String(e)}),e}},s)};return{id:e,type:"ggml-stt",info:n.info,queue:r.queue,initContext:async(e={})=>{const{onProgress:t}=e,n=await Fr(r,t);return{modelInfo:n.modelInfo?{...n.modelInfo}:null,runtime:{...r.plan.info.runtime},download:{...r.plan.info.download}}},transcribe:async(e={})=>o(e),transcribeData:async(e={})=>o(e),releaseContext:async(e={})=>{const t=r.contextRecord;return!!t&&Ur(r,t,Boolean(e.force))},finalize:async()=>{r.contextRecord&&await Ur(r,r.contextRecord,!0)},getStatus:()=>({id:r.id,type:r.type,repoId:r.plan.info.model?.repoId||null,quantization:r.plan.info.model?.quantization||null,variant:r.plan.info.runtime?.variant||null,hasContext:Boolean(r.contextRecord?.context),contextRefCount:r.contextRecord?.refCount||0,queueStatus:r.queue.getStatus()})}}const Dr=e=>{const t=Cr(e);return t.model.repo_id||t.model.repository||t.model.model||null},Rr=e=>e?"number"==typeof e.score&&Number.isFinite(e.score)?Number(e.score):Lt(e):0;async function Lr(e,t=null,n={}){if("ggml-llm"===e)return fr(t,n);if("ggml-stt"===e)return async function(e=null,t={}){const{threshold:n=1.1,includeBreakdown:r=!1,config:o,...s}=t;let l=null,c=null,u=null;if(o)try{const e=Cr(o),t=await Nr(e);l=t.size??null,({processingBufferBytes:c}=rn({modelBytes:l})),u=t.quantization||null}catch(e){}const d=null!=o?.backend?.gpu_memory_fraction?Math.min(1,Math.max(0,Number(o.backend.gpu_memory_fraction))):void 0,p=null!=o?.backend?.cpu_memory_fraction?Math.min(1,Math.max(0,Number(o.backend.cpu_memory_fraction))):void 0,m=await en({...s,platform:process.platform,totalMemoryInBytes:Pt.totalmem(),backend:"ggml-stt",includeBreakdown:r,gpuMemoryFraction:d,cpuMemoryFraction:p,dependencies:{getBackendDevicesInfo:a,isLibVariantAvailable:i},modelBytes:l,kvCacheBytes:c}),f=m.selected,h=Rr(f);f&&(f.modelBytes=l||null,f.processingBytes=c||null,f.quantization=u||null);let y=null,g=null;if(e){const t=Rr(e);g={...e,score:t};let r="buttress",o="buttress-higher-score";if(m.ok)if(t||0===t)if(e.fit&&f?.fit){const a=e.fit.fitsInGpu||e.fit.fitsInCpu,i=f.fit.fitsInGpu||f.fit.fitsInCpu;a&&!i?(r="local",o="client-fits-in-memory"):i&&!a?(r="buttress",o="buttress-fits-in-memory"):t>h*n?(r="local",o="client-better"):h>t*n?(r="buttress",o="buttress-better"):(r="either",o="comparable-scores")}else t>h*n?(r="local",o="client-better"):h>t*n?(r="buttress",o="buttress-better"):(r="either",o="comparable-scores");else r="buttress",o="missing-client-score";else r="local",o="buttress-unavailable";y={buttressScore:h,clientScore:t,threshold:n,recommendation:r,reason:o}}m.ok||y||(y={buttressScore:h,clientScore:e?.score??null,threshold:n,recommendation:"local",reason:"buttress-unavailable"});let b=null;return o&&(b={repoId:o.model?.repo_id||null,quantization:o.model?.quantization||null,filename:o.model?.filename||null}),{type:"ggml-stt",timestamp:(new Date).toISOString(),buttress:m,client:g,comparison:y,modelConfig:b}}(t,n);throw new Error(`Unknown backend type: ${e}`)}const Gr=t("node:fs"),Hr=t("node:child_process"),Jr=JSON.parse('{"UU":"@fugood/buttress-backend-core","rE":"2.23.0-beta.32"}'),Vr=e=>{if(!e)return{repoId:null,filename:null};const[t,n]=e.split(":");return{repoId:t,filename:n||null}};async function Kr({modelIds:e=[],defaultConfig:t=null}={}){const n=[];console.log(`${Jr.UU} v${Jr.rE}`),console.log("Generating model capabilities comparison...\n"),n.push(`${Jr.UU} v${Jr.rE}`),n.push("## Model Capabilities Comparison\n"),e&&0!==e.length||(console.error("Error: No model IDs provided"),process.exit(1));try{const r=(e={},t={})=>{const n=Array.isArray(e)?[...e]:{...e};return Object.entries(t||{}).forEach(([e,t])=>{t&&"object"==typeof t&&!Array.isArray(t)?n[e]=r(n[e]||{},t):n[e]=t}),n},o=t||{},{server:a,generators:i=[],...s}=o,l=e=>r(JSON.parse(JSON.stringify(s)),e||{}),u=e=>{if(Array.isArray(i)&&i.length>0){const t=i.filter(e=>"ggml-llm"===e?.type);if(t.length>0&&e){const n=t.find(t=>t.model?.repo_id===e);if(n)return l(n)}}return Object.keys(s).length>0?l({}):null},d=[];for(let t=0;t<e.length;t+=1){const n=e[t];console.log(`[${t+1}/${e.length}] Analyzing ${n}...`);let r=u(n);r={...r||{},model:{...s.runtime,...r?.model||{},repo_id:n}};const o=await Lr("ggml-llm",null,{config:r,includeBreakdown:!0});d.push({modelId:n,capabilities:o,modelInfo:o.buttress?.selected||null,modelConfig:o.modelConfig||null})}const p=e=>e?(e/1024/1024/1024).toFixed(2):"N/A",m=e=>e?"✅":"🚫";n.push("| Model ID | Quantization | Size (GB) | Context Size | KV Cache Size (GB) | Total Required Memory (GB) | Fits GPU (Full) | Fits CPU (Full) |"),n.push("|----------|--------------|-----------|--------------|--------------------|-----------------------------|-----------------|-----------------|"),d.forEach(({modelId:e,modelInfo:t,modelConfig:r})=>{const o=t?.quantization?.name?.toUpperCase()||"N/A",a=p(t?.modelBytes),i=r?.nCtx||t?.kvInfo?.nCtxTrain||"N/A",s=un(t),l=Number(i),c=t?.kvCacheBytes||(s&&Number.isFinite(l)&&l>0?s(l):s&&s(t?.kvInfo?.nCtxTrain||0))||null,u=p(c),d=p(t?.modelBytes&&c?t.modelBytes+c:t?.fit?.totalRequiredBytes),f=m(t?.fit?.fitsInGpu),h=m(t?.fit?.fitsInCpu);n.push(`| ${e} | ${o} | ${a} | ${i} | ${u} | ${d} | ${f} | ${h} |`);if((null!=t?.memoryLimitedCtx||null!=t?.limitedFit)&&(!t?.fit?.fitsInGpu||!t?.fit?.fitsInCpu)){const e=t?.memoryLimitedCtx||i,r=Number(e),o=t?.limitedKvCacheBytes||s&&Number.isFinite(r)&&r>0&&s(r)||null,l=p(o),c=p(t?.modelBytes&&o?t.modelBytes+o:t?.limitedFit?.totalRequiredBytes),f=m(t?.limitedFit?.fitsInGpu),h=m(t?.limitedFit?.fitsInCpu);(e!==i||l!==u||c!==d)&&n.push(`| ↳ Limited | - | ${a} | ${e} | ${l} | ${c} | ${f} | ${h} |`)}}),n.push("\n---"),n.push("\n### System Information");let f=null;if("win32"!==process.platform)try{f=(0,Hr.execSync)("uname -a",{encoding:"utf8"}).trim()}catch{}if(f?n.push(`- **System:** ${f}`):(n.push(`- **Hostname:** ${Pt.hostname()}`),n.push(`- **OS:** ${Pt.type()} ${Pt.release()}`)),n.push(`- **Platform:** ${process.platform}`),n.push(`- **CPU Cores:** ${Pt.cpus().length}`),n.push(`- **Total System Memory:** ${(Pt.totalmem()/1024/1024/1024).toFixed(2)} GB`),d.length>0){const e=d[0].capabilities,t=e.buttress?.selected;if(t){const e=t.cpuTotalBytes>0?(t.cpuUsableBytes/t.cpuTotalBytes*100).toFixed(0):0;if(n.push(`- **Usable CPU Memory:** ${(t.cpuUsableBytes/1024/1024/1024).toFixed(2)} GB (${e}% of ${(t.cpuTotalBytes/1024/1024/1024).toFixed(2)} GB)`),t.hasGpu){const e=t.devices.filter(e=>"gpu"===e.type);if(e.length>0){const r=e[0];n.push(`- **GPU Backend:** ${r.backend}`),n.push(`- **GPU Name:** ${r.deviceName}`),n.push(`- **GPU Total Memory:** ${(r.maxMemorySize/1024/1024/1024).toFixed(2)} GB`);const o=t.gpuTotalBytes>0?(t.gpuUsableBytes/t.gpuTotalBytes*100).toFixed(0):0;n.push(`- **GPU Usable Memory:** ${(t.gpuUsableBytes/1024/1024/1024).toFixed(2)} GB (${o}% of ${(t.gpuTotalBytes/1024/1024/1024).toFixed(2)} GB)`)}}else n.push("- **GPU:** Not available")}}n.push("\n### Command Used");const h=process.argv.slice(2).join(" ");if(n.push(`\`\`\`bash\n${process.argv[0]} ${process.argv[1]} ${h}\n\`\`\``),n.push("\n### Package Information"),n.push(`- **Name:** ${Jr.UU}`),n.push(`- **Version:** ${Jr.rE}`),Jr.description&&n.push(`- **Description:** ${Jr.description}`),t&&Object.keys(t).length>0){n.push("\n### Configuration"),n.push("<details>"),n.push("<summary>Click to expand TOML configuration</summary>"),n.push("\n```toml");try{const e=c.stringify(t);n.push(e)}catch(e){n.push("# Error serializing config"),n.push(JSON.stringify(t,null,2))}n.push("```"),n.push("</details>")}const y=`ggml-llm-model-capabilities-${(new Date).toISOString().replace(/[.:]/g,"-").split("T")[0]}.md`,b=g.join(process.cwd(),y);Gr.writeFileSync(b,n.join("\n"),"utf8"),console.log(`\nModel capabilities table saved to: ${b}`),process.exit(0)}catch(e){console.error("Failed to generate model table:",e.message),process.exit(1)}}async function Yr({modelId:e=null,defaultConfig:t=null}={}){console.log(`${Jr.UU} v${Jr.rE}`),console.log("Testing capabilities for backend: ggml-llm"),e&&console.log(`Model: ${e}`),console.log("--------------------------------");try{const n=t||{},{server:r,generators:o=[],...a}=n,i=(e={},t={})=>{const n=Array.isArray(e)?[...e]:{...e};return Object.entries(t||{}).forEach(([e,t])=>{t&&"object"==typeof t&&!Array.isArray(t)?n[e]=i(n[e]||{},t):n[e]=t}),n},s=e=>i(JSON.parse(JSON.stringify(a)),e||{});let l=(e=>{if(Array.isArray(o)&&o.length>0){const t=o.filter(e=>"ggml-llm"===e?.type);if(t.length>0&&e){const n=t.find(t=>t.model?.repo_id===e);if(n)return s(n)}}return Object.keys(a).length>0?s({}):null})(e);e&&(l={...l||{},model:{...l?.model||{},repo_id:e}});const c=await Lr("ggml-llm",null,{config:l,includeBreakdown:!0}),u=c.buttress?.selected||null,d=c.modelConfig||null;if(e||d?.repoId){console.log("\n=== Model Information ===");const t=e||d?.repoId;console.log(`Repository ID: ${t}`),d?.quantization&&console.log(`Quantization: ${d.quantization}`),d?.nCtx&&console.log(`Context Length: ${d.nCtx}`),u?.quantization?.name&&console.log(`Model Quantization: ${u.quantization.name.toUpperCase()}`);const n=d?.cache_type_k||"f16",r=d?.cache_type_v||"f16";if(console.log(`KV Cache Type: K=${n}, V=${r}`),u?.modelBytes&&u?.kvCacheBytes){if(console.log(`Model Size: ${(u.modelBytes/1024/1024/1024).toFixed(2)} GB`),u.kvInfo?console.log(`KV Cache Size: ${(u.kvCacheBytes/1024/1024/1024).toFixed(2)} GB (KV info: ${JSON.stringify(u.kvInfo)})`):console.log(`KV Cache Size: ${(u.kvCacheBytes/1024/1024/1024).toFixed(2)} GB`),console.log(`Total Required Memory: ${((u.modelBytes+u.kvCacheBytes)/1024/1024/1024).toFixed(2)} GB`),null!=u.memoryLimitedCtx){const e=u.memoryLimitedCtx,t=u.kvInfo?.nCtxTrain;t?console.log(`\nMemory-Limited Context: ${e} (Train: ${t})`):console.log(`\nMemory-Limited Context: ${e}`),null!=u.limitedKvCacheBytes&&console.log(`Limited KV Cache Size: ${(u.limitedKvCacheBytes/1024/1024/1024).toFixed(2)} GB`)}}else if(c.buttress?.selected?.fit){const{totalRequiredBytes:e}=c.buttress.selected.fit;console.log(`Total Required Memory: ${(e/1024/1024/1024).toFixed(2)} GB`)}}if(c.buttress?.selected){const{selected:e}=c.buttress;console.log("\n=== Hardware Information ===");let t=null;if("win32"!==process.platform)try{t=(0,Hr.execSync)("uname -a",{encoding:"utf8"}).trim()}catch{}t?console.log(`System: ${t}`):(console.log(`Hostname: ${Pt.hostname()}`),console.log(`OS: ${Pt.type()} ${Pt.release()}`)),console.log(`Platform: ${e.platform}`),console.log(`CPU Cores: ${Pt.cpus().length}`),console.log(`Total System Memory: ${(Pt.totalmem()/1024/1024/1024).toFixed(2)} GB`);const n=e.cpuTotalBytes>0?(e.cpuUsableBytes/e.cpuTotalBytes*100).toFixed(0):0;if(console.log(`Usable CPU Memory: ${(e.cpuUsableBytes/1024/1024/1024).toFixed(2)} GB (${n}% of ${(e.cpuTotalBytes/1024/1024/1024).toFixed(2)} GB)`),e.hasGpu){console.log("\n--- GPU Details ---");e.devices.filter(e=>"gpu"===e.type).forEach(t=>{console.log(`GPU Backend: ${t.backend}`),console.log(`GPU Name: ${t.deviceName}`),console.log(`GPU Total Memory: ${(t.maxMemorySize/1024/1024/1024).toFixed(2)} GB`);const n=e.gpuTotalBytes>0?(e.gpuUsableBytes/e.gpuTotalBytes*100).toFixed(0):0;console.log(`GPU Usable Memory: ${(e.gpuUsableBytes/1024/1024/1024).toFixed(2)} GB (${n}% of ${(e.gpuTotalBytes/1024/1024/1024).toFixed(2)} GB)`),t.metadata&&(t.metadata.hasBFloat16&&console.log("Supports BFloat16: Yes"),t.metadata.hasUnifiedMemory&&console.log("Unified Memory: Yes"))})}else console.log("GPU: Not available");console.log(`\nBackend Variant: ${e.variant}`),console.log(`Performance Score: ${e.score}`),e.fit&&(console.log("\n--- Model Fit Analysis ---"),console.log("Fits in GPU: "+(e.fit.fitsInGpu?"Yes":"No")),console.log("Fits in CPU: "+(e.fit.fitsInCpu?"Yes":"No")),console.log(`Limiting Factor: ${e.fit.limiting}`),e.limitedFit&&(console.log("\n--- Memory-Limited Fit Analysis ---"),console.log(`Limited Total Required: ${(e.limitedFit.totalRequiredBytes/1024/1024/1024).toFixed(2)} GB`),console.log("Fits in GPU (Limited): "+(e.limitedFit.fitsInGpu?"Yes":"No")),console.log("Fits in CPU (Limited): "+(e.limitedFit.fitsInCpu?"Yes":"No")),console.log(`Limiting Factor (Limited): ${e.limitedFit.limiting}`)))}console.log("\n=== Full Capabilities JSON ==="),console.log(JSON.stringify(c,null,2)),process.exit(0)}catch(e){console.error("Failed to get capabilities:",e.message),process.exit(1)}}async function Wr({modelIds:e=[],defaultConfig:t=null}={}){const n=[];console.log(`${Jr.UU} v${Jr.rE}`),console.log("Generating STT model capabilities comparison...\n"),n.push(`${Jr.UU} v${Jr.rE}`),n.push("## STT Model Capabilities Comparison\n"),e&&0!==e.length||(console.error("Error: No model IDs provided"),process.exit(1));try{const r=(e={},t={})=>{const n=Array.isArray(e)?[...e]:{...e};return Object.entries(t||{}).forEach(([e,t])=>{t&&"object"==typeof t&&!Array.isArray(t)?n[e]=r(n[e]||{},t):n[e]=t}),n},o=t||{},{server:a,generators:i=[],...s}=o,l=e=>r(JSON.parse(JSON.stringify(s)),e||{}),u=e=>{if(Array.isArray(i)&&i.length>0){const t=i.filter(e=>"ggml-stt"===e?.type);if(t.length>0&&e){const n=t.find(t=>t.model?.repo_id===e);if(n)return l(n)}}return Object.keys(s).length>0?l({}):null},d=[];for(let t=0;t<e.length;t+=1){const n=e[t],{repoId:r,filename:o}=Vr(n);console.log(`[${t+1}/${e.length}] Analyzing ${n}...`);let a=u(r);a={...a||{},model:{...a?.model||{},repo_id:r,...o&&{filename:o}}};const i=await Lr("ggml-stt",null,{config:a,includeBreakdown:!0});d.push({modelId:n,repoId:r,filename:o,capabilities:i,modelInfo:i.buttress?.selected||null,modelConfig:i.modelConfig||null})}const p=e=>e?(e/1024/1024).toFixed(1):"N/A",m=e=>e?"✅":"🚫";n.push("| Model | Size (MB) | Processing Buffer (MB) | Total Required (MB) | Fits GPU | Fits CPU |"),n.push("|-------|-----------|------------------------|---------------------|----------|----------|"),d.forEach(({modelId:e,modelInfo:t})=>{const r=p(t?.modelBytes),o=p(t?.processingBytes||t?.kvCacheBytes),a=p(t?.fit?.totalRequiredBytes),i=m(t?.fit?.fitsInGpu),s=m(t?.fit?.fitsInCpu);n.push(`| ${e} | ${r} | ${o} | ${a} | ${i} | ${s} |`)}),n.push("\n---"),n.push("\n### System Information");let f=null;if("win32"!==process.platform)try{f=(0,Hr.execSync)("uname -a",{encoding:"utf8"}).trim()}catch{}if(f?n.push(`- **System:** ${f}`):(n.push(`- **Hostname:** ${Pt.hostname()}`),n.push(`- **OS:** ${Pt.type()} ${Pt.release()}`)),n.push(`- **Platform:** ${process.platform}`),n.push(`- **CPU Cores:** ${Pt.cpus().length}`),n.push(`- **Total System Memory:** ${(Pt.totalmem()/1024/1024/1024).toFixed(2)} GB`),d.length>0){const e=d[0].capabilities,t=e.buttress?.selected;if(t){const e=t.cpuTotalBytes>0?(t.cpuUsableBytes/t.cpuTotalBytes*100).toFixed(0):0;if(n.push(`- **Usable CPU Memory:** ${(t.cpuUsableBytes/1024/1024/1024).toFixed(2)} GB (${e}% of ${(t.cpuTotalBytes/1024/1024/1024).toFixed(2)} GB)`),t.hasGpu){const e=t.devices.filter(e=>"gpu"===e.type);if(e.length>0){const r=e[0];n.push(`- **GPU Backend:** ${r.backend}`),n.push(`- **GPU Name:** ${r.deviceName}`),n.push(`- **GPU Total Memory:** ${(r.maxMemorySize/1024/1024/1024).toFixed(2)} GB`);const o=t.gpuTotalBytes>0?(t.gpuUsableBytes/t.gpuTotalBytes*100).toFixed(0):0;n.push(`- **GPU Usable Memory:** ${(t.gpuUsableBytes/1024/1024/1024).toFixed(2)} GB (${o}% of ${(t.gpuTotalBytes/1024/1024/1024).toFixed(2)} GB)`)}}else n.push("- **GPU:** Not available")}}n.push("\n### Command Used");const h=process.argv.slice(2).join(" ");if(n.push(`\`\`\`bash\n${process.argv[0]} ${process.argv[1]} ${h}\n\`\`\``),n.push("\n### Package Information"),n.push(`- **Name:** ${Jr.UU}`),n.push(`- **Version:** ${Jr.rE}`),Jr.description&&n.push(`- **Description:** ${Jr.description}`),t&&Object.keys(t).length>0){n.push("\n### Configuration"),n.push("<details>"),n.push("<summary>Click to expand TOML configuration</summary>"),n.push("\n```toml");try{const e=c.stringify(t);n.push(e)}catch(e){n.push("# Error serializing config"),n.push(JSON.stringify(t,null,2))}n.push("```"),n.push("</details>")}const y=`ggml-stt-model-capabilities-${(new Date).toISOString().replace(/[.:]/g,"-").split("T")[0]}.md`,b=g.join(process.cwd(),y);Gr.writeFileSync(b,n.join("\n"),"utf8"),console.log(`\nSTT model capabilities table saved to: ${b}`),process.exit(0)}catch(e){console.error("Failed to generate STT model table:",e.message),process.exit(1)}}async function Qr({modelId:e=null,defaultConfig:t=null}={}){console.log(`${Jr.UU} v${Jr.rE}`),console.log("Testing capabilities for backend: ggml-stt"),e&&console.log(`Model: ${e}`),console.log("--------------------------------");try{const{repoId:n,filename:r}=Vr(e),o=t||{},{server:a,generators:i=[],...s}=o,l=(e={},t={})=>{const n=Array.isArray(e)?[...e]:{...e};return Object.entries(t||{}).forEach(([e,t])=>{t&&"object"==typeof t&&!Array.isArray(t)?n[e]=l(n[e]||{},t):n[e]=t}),n},c=e=>l(JSON.parse(JSON.stringify(s)),e||{});let u=(e=>{if(Array.isArray(i)&&i.length>0){const t=i.filter(e=>"ggml-stt"===e?.type);if(t.length>0&&e){const n=t.find(t=>t.model?.repo_id===e);if(n)return c(n)}}return Object.keys(s).length>0?c({}):null})(n);n&&(u={...u||{},model:{...s.runtime,...u?.model||{},repo_id:n,...r&&{filename:r}}});const d=await Lr("ggml-stt",null,{config:u,includeBreakdown:!0}),p=d.buttress?.selected||null,m=d.modelConfig||null;if(n||m?.repoId){console.log("\n=== Model Information ===");const e=n||m?.repoId;console.log(`Repository ID: ${e}`),r&&console.log(`Filename: ${r}`),p?.modelBytes&&console.log(`Model Size: ${(p.modelBytes/1024/1024).toFixed(1)} MB`);const t=p?.processingBytes||p?.kvCacheBytes;if(t&&console.log(`Processing Buffer: ${(t/1024/1024).toFixed(1)} MB`),p?.modelBytes&&t)console.log(`Total Required Memory: ${((p.modelBytes+t)/1024/1024).toFixed(1)} MB`);else if(d.buttress?.selected?.fit){const{totalRequiredBytes:e}=d.buttress.selected.fit;console.log(`Total Required Memory: ${(e/1024/1024).toFixed(1)} MB`)}}if(d.buttress?.selected){const{selected:e}=d.buttress;console.log("\n=== Hardware Information ===");let t=null;if("win32"!==process.platform)try{t=(0,Hr.execSync)("uname -a",{encoding:"utf8"}).trim()}catch{}t?console.log(`System: ${t}`):(console.log(`Hostname: ${Pt.hostname()}`),console.log(`OS: ${Pt.type()} ${Pt.release()}`)),console.log(`Platform: ${e.platform}`),console.log(`CPU Cores: ${Pt.cpus().length}`),console.log(`Total System Memory: ${(Pt.totalmem()/1024/1024/1024).toFixed(2)} GB`);const n=e.cpuTotalBytes>0?(e.cpuUsableBytes/e.cpuTotalBytes*100).toFixed(0):0;if(console.log(`Usable CPU Memory: ${(e.cpuUsableBytes/1024/1024/1024).toFixed(2)} GB (${n}% of ${(e.cpuTotalBytes/1024/1024/1024).toFixed(2)} GB)`),e.hasGpu){console.log("\n--- GPU Details ---");e.devices.filter(e=>"gpu"===e.type).forEach(t=>{console.log(`GPU Backend: ${t.backend}`),console.log(`GPU Name: ${t.deviceName}`),console.log(`GPU Total Memory: ${(t.maxMemorySize/1024/1024/1024).toFixed(2)} GB`);const n=e.gpuTotalBytes>0?(e.gpuUsableBytes/e.gpuTotalBytes*100).toFixed(0):0;console.log(`GPU Usable Memory: ${(e.gpuUsableBytes/1024/1024/1024).toFixed(2)} GB (${n}% of ${(e.gpuTotalBytes/1024/1024/1024).toFixed(2)} GB)`),t.metadata&&(t.metadata.hasBFloat16&&console.log("Supports BFloat16: Yes"),t.metadata.hasUnifiedMemory&&console.log("Unified Memory: Yes"))})}else console.log("GPU: Not available");console.log(`\nBackend Variant: ${e.variant}`),console.log(`Performance Score: ${e.score}`),e.fit&&(console.log("\n--- Model Fit Analysis ---"),console.log("Fits in GPU: "+(e.fit.fitsInGpu?"Yes":"No")),console.log("Fits in CPU: "+(e.fit.fitsInCpu?"Yes":"No")),console.log(`Limiting Factor: ${e.fit.limiting}`))}console.log("\n=== Full Capabilities JSON ==="),console.log(JSON.stringify(d,null,2)),process.exit(0)}catch(e){console.error("Failed to get capabilities:",e.message),process.exit(1)}}const Zr=new Map,Xr=(e,t)=>{const n=(e=>{const t=Zr.get(e);if(!t)throw new Error(`Unknown generator id "${e}"`);return t})(e);if(n.type!==t)throw new Error(`Generator "${e}" does not support ${t} backend`);return n.instance};async function eo(e,t){const n={"ggml-llm":{create:dr,getId:pr},"ggml-stt":{create:jr,getId:Dr}}[e];if(!n)throw new Error(`Unsupported backend type: ${e}`);const r=n.getId(t);if(!r)throw new Error("Buttress generator config missing repo identifier");const o=`${e}:${r}`,a=Zr.get(o);if(a)return a.refCount+=1,{id:a.id,info:a.instance.info};const i=await n.create(o,t),s={id:o,type:i.type,instance:i,refCount:1};return Zr.set(o,s),{id:o,info:i.info}}async function to(e){const t=Zr.get(e);return!!t&&(t.refCount-=1,t.refCount<=0&&(await t.instance.finalize(),Zr.delete(e)),!0)}const no={initContext:async(e,t)=>Xr(e,"ggml-llm").initContext(t),completion:async(e,t)=>Xr(e,"ggml-llm").completion(t),tokenize:async(e,t)=>Xr(e,"ggml-llm").tokenize(t),detokenize:async(e,t)=>Xr(e,"ggml-llm").detokenize(t),applyChatTemplate:async(e,t)=>Xr(e,"ggml-llm").applyChatTemplate(t),releaseContext:async(e,t)=>Xr(e,"ggml-llm").releaseContext(t)},ro={initContext:async(e,t)=>Xr(e,"ggml-stt").initContext(t),transcribe:async(e,t)=>Xr(e,"ggml-stt").transcribe(t),transcribeData:async(e,t)=>Xr(e,"ggml-stt").transcribeData(t),releaseContext:async(e,t)=>Xr(e,"ggml-stt").releaseContext(t)};function oo(e,t){return"ggml-llm"===e?pr(t):"ggml-stt"===e?Dr(t):null}const ao={getFullStatus:()=>function(e){return{timestamp:(new Date).toISOString(),ggmlLlm:Sn(e),ggmlStt:kn(e)}}(Zr),getGgmlLlmStatus:()=>Sn(Zr),getGgmlSttStatus:()=>kn(Zr),subscribeToStatus:xn,subscribeToStatusWithId:function(e){return _n+=1,{subscriberId:_n,unsubscribe:xn(e)}},llmStatusTracker:vn,sttStatusTracker:wn,statusEmitter:yn},{version:io,name:so}={version:"2.23.0-beta.32",name:"@fugood/buttress-server-poc"},lo=async()=>{const e=`https://registry.npmjs.org/${so}/latest`;try{const t=new AbortController,n=setTimeout(()=>t.abort(),3e3),r=await fetch(e,{headers:{Accept:"application/json"},signal:t.signal});if(clearTimeout(n),!r.ok)return null;return(await r.json()).version||null}catch{return null}},co=(e,t)=>{if(!t)return!1;const n=e.split(/[.-]/),r=t.split(/[.-]/);for(let e=0;e<Math.max(n.length,r.length);e+=1){const t=parseInt(n[e])||0,o=parseInt(r[e])||0;if(o>t)return!0;if(o<t)return!1}return!1},uo=async()=>{try{const e=await lo();e&&co(io,e)&&(e=>{console.log(""),console.log("╭─────────────────────────────────────────────────╮"),console.log("│ Update available! %s → %s",io.padEnd(12),e.padEnd(12),"│"),console.log("│ │"),console.log("│ Run to upgrade: │"),console.log("│ npm install -g %s │",so.padEnd(27)),console.log("╰─────────────────────────────────────────────────╯"),console.log("")})(e)}catch(e){}},po=n.create(),mo=e=>{const{status:t}=e;return po.router({getStatus:po.procedure.query(()=>(console.log("[Status] Get full status"),t.getFullStatus())),getLlmStatus:po.procedure.query(()=>(console.log("[Status] Get LLM status"),t.getGgmlLlmStatus())),getSttStatus:po.procedure.query(()=>(console.log("[Status] Get STT status"),t.getGgmlSttStatus())),subscribe:po.procedure.subscription(()=>te(n=>{console.log("[Status] Client subscribed to status changes");const{generatorRegistry:r}=e,o=[],a=(e,r)=>{if("ggml-llm"===r.type&&r.instance?.subscribeParallelStatus){const a=r.instance.subscribeParallelStatus(r=>{n.next({type:"parallelStatus",generatorId:e,data:r,fullStatus:t.getFullStatus()})});o.push({id:e,sub:a})}};Array.from(r.entries()).forEach(([e,t])=>{a(e,t)}),n.next({type:"initial",data:t.getFullStatus()});const i=t.subscribeToStatus(e=>{if("modelLoad"===e.type&&e.entry?.generatorId){const t=r.get(e.entry.generatorId);t&&a(e.entry.generatorId,t)}n.next({type:"change",data:e,fullStatus:t.getFullStatus()})});return()=>{console.log("[Status] Client unsubscribed from status changes"),o.forEach(({sub:e})=>e?.remove?.()),i()}}))})},fo=new TextDecoder,ho=(e,t,n)=>{e.writeHead(t,{"Content-Type":"application/json","Access-Control-Allow-Origin":"*"}),e.end(JSON.stringify(n))},yo=(e,t)=>{e.write(`data: ${JSON.stringify(t)}\n\n`)},go=(e,t)=>{const n=e.split("\n\n"),r=n.pop()||"";return n.forEach(e=>{if(!e.trim())return;const n=e.match(/^event:\s*(.+)$/m),r=e.match(/^data:\s*(.+)$/m);if(!n||!r)return;const o=n[1];let a;try{a=JSON.parse(r[1])}catch{return}t(o,a)}),r},bo=({backend:e,defaultConfig:t})=>{const n=new Map,r=async(e,t)=>{const n=e.getReader();let r="";try{let e=!1;for(;!e;){const{value:o,done:a}=await n.read();if(e=a,e)break;r+=fo.decode(o,{stream:!0}),r=go(r,(e,n)=>{"token"===e&&t.onToken?t.onToken(n):"result"===e&&t.onResult?t.onResult(n):"error"===e&&t.onError&&t.onError(n)})}}finally{n.cancel().catch(()=>{})}},o=async(o,a)=>{let i;try{i=await(e=>new Promise((t,n)=>{let r="";e.on("data",e=>{r+=e.toString()}),e.on("end",()=>{try{t(r?JSON.parse(r):{})}catch(e){n(new Error("Invalid JSON body"))}}),e.on("error",n)}))(o)}catch(e){return void ho(a,400,{error:{message:e.message,type:"invalid_request_error"}})}const{messages:s=[],stream:l=!1,model:c,tools:u,temperature:d,stop:p}=i,m=i.top_p,f=i.max_tokens,h=i.presence_penalty,y=i.frequency_penalty,g=i.tool_choice,b=i.stream_options;if(s&&0!==s.length)try{const o=await(async r=>{const o=(t?.generators||[]).filter(e=>"ggml-llm"===e.type);if(0===o.length)throw new Error('No ggml-llm generator configured. Add a [[generators]] with type = "ggml-llm" to your config.');let a=o[0],i=r||a.model?.repo_id;if(r){const e=o.find(e=>e.model?.repo_id===r);e&&(a=e)}else i=a.model?.repo_id;const s=i,l=n.get(s);if(l?.initialized)return l;const{generators:c,server:u,...d}=t||{},p={...d,...a,model:{...a.model,repo_id:i}};console.log(`[OpenAI] Creating generator for ${s}`);const{id:m}=await e.startGenerator("ggml-llm",p),f={id:m,config:p,repoId:i,initialized:!1};return n.set(s,f),await e.ggmlLlm.initContext(m,{}),f.initialized=!0,console.log(`[OpenAI] Generator ready: ${s}`),f})(c),i=`chatcmpl-${Date.now()}-${Math.random().toString(36).slice(2,9)}`,v=Math.floor(Date.now()/1e3),w=o.repoId||"ggml-llm",_={reasoning_format:"auto",messages:s,jinja:!0,add_generation_prompt:!0};null!=d&&(_.temperature=d),null!=m&&(_.top_p=m),null!=f&&(_.n_predict=f),null!=p&&(_.stop=Array.isArray(p)?p:[p]),null!=h&&(_.presence_penalty=h),null!=y&&(_.frequency_penalty=y),null!=u&&(_.tools=u),null!=g&&(_.tool_choice=g);const x=await e.ggmlLlm.completion(o.id,{options:_});if(l){a.writeHead(200,{"Content-Type":"text/event-stream","Cache-Control":"no-cache",Connection:"keep-alive","Access-Control-Allow-Origin":"*"});const e=!0===b?.include_usage;let t="",n="";const o=new Map,s=new Map;await r(x,{onToken:e=>{const r={};if(null!=e.content){const n=e.content;n.length>t.length&&(r.content=n.slice(t.length),t=n)}if(null!=e.reasoning_content){const t=e.reasoning_content;t.length>n.length&&(r.reasoning_content=t.slice(n.length),n=t)}if(null!=e.tool_calls&&e.tool_calls.length>0){const t=[];e.tool_calls.forEach((e,n)=>{const r={index:n};s.has(n)||(s.set(n,e.id||`call_${i}_${n}`),r.id=s.get(n),r.type="function");const a=e.function?.arguments||"",l=o.get(n)||"",c={};!o.has(n)&&e.function?.name&&(c.name=e.function.name),a.length>l.length&&(c.arguments=a.slice(l.length),o.set(n,a)),Object.keys(c).length>0?(r.function=c,t.push(r)):r.id&&(r.function={name:e.function?.name||"",arguments:""},t.push(r))}),t.length>0&&(r.tool_calls=t)}Object.keys(r).length>0&&yo(a,{id:i,object:"chat.completion.chunk",created:v,model:w,choices:[{index:0,delta:r,finish_reason:null}]})},onResult:t=>{const n={id:i,object:"chat.completion.chunk",created:v,model:w,choices:[{index:0,delta:{},finish_reason:t.interrupted?"length":"stop"}]};e&&(n.usage={prompt_tokens:t.prompt_tokens||t.promptTokens||0,completion_tokens:t.tokens_predicted||t.tokensPredicted||0,total_tokens:(t.prompt_tokens||t.promptTokens||0)+(t.tokens_predicted||t.tokensPredicted||0)}),yo(a,n)},onError:e=>{yo(a,{error:{message:e.message,type:"server_error"}})}}),a.write("data: [DONE]\n\n"),a.end()}else{let e="",t=null,n=null,o="stop",s={prompt_tokens:0,completion_tokens:0,total_tokens:0};await r(x,{onToken:t=>{null!=t.content?e+=t.content:null!=t.token&&(e+=t.token)},onResult:r=>{r.text?e=r.text:r.content&&(e=r.content),r.reasoning_content&&(t=r.reasoning_content),r.tool_calls&&r.tool_calls.length>0?(n=r.tool_calls.map((e,t)=>({id:e.id||`call_${i}_${t}`,type:"function",function:{name:e.function?.name||"",arguments:e.function?.arguments||""}})),o="tool_calls"):o=r.interrupted?"length":"stop",s={prompt_tokens:r.prompt_tokens||r.promptTokens||0,completion_tokens:r.tokens_predicted||r.tokensPredicted||0,total_tokens:(r.prompt_tokens||r.promptTokens||0)+(r.tokens_predicted||r.tokensPredicted||0)}},onError:e=>{throw new Error(e.message)}});const l={role:"assistant",content:e||null};t&&(l.reasoning_content=t),n&&(l.tool_calls=n),ho(a,200,{id:i,object:"chat.completion",created:v,model:w,choices:[{index:0,message:l,finish_reason:o}],usage:s})}}catch(e){console.error("[OpenAI] Chat completion error:",e),ho(a,500,{error:{message:e.message||"Internal server error",type:"server_error"}})}else ho(a,400,{error:{message:"messages is required and must not be empty",type:"invalid_request_error"}})};return async(e,n,r)=>{if("OPTIONS"===e.method)return n.writeHead(204,{"Access-Control-Allow-Origin":"*","Access-Control-Allow-Methods":"GET, POST, OPTIONS","Access-Control-Allow-Headers":"Content-Type, Authorization","Access-Control-Max-Age":"86400"}),n.end(),!0;const{pathname:a}=r;return"/oai-compat/v1/models"===a&&"GET"===e.method?(await(async(e,n)=>{const r=(t?.generators||[]).filter(e=>"ggml-llm"===e.type).map(e=>({id:e.model?.repo_id||"ggml-llm",object:"model",created:Math.floor(Date.now()/1e3),owned_by:"local"}));0===r.length&&r.push({id:"ggml-llm",object:"model",created:Math.floor(Date.now()/1e3),owned_by:"local"}),ho(n,200,{object:"list",data:r})})(0,n),!0):"/oai-compat/v1/chat/completions"===a&&"POST"===e.method&&(await o(e,n),!0)}},vo=g.dirname((0,b.fileURLToPath)(import.meta.url)),wo=new TextDecoder,_o=n.context().create({sse:{ping:{enabled:!0,intervalMs:3e4},maxDurationMs:18e5,client:{reconnectAfterInactivityMs:3e5}}}),xo=r.record(r.any()).optional(),So=r.object({id:r.string()}),ko=r.object({id:r.string(),property:r.any().optional()}),Co=new Map,Eo=r.object({type:r.string(),config:xo}),Mo=r.object({type:r.string().optional().default("ggml-llm"),config:xo,currentClientCapabilities:r.record(r.any()).optional(),options:r.record(r.any()).optional()}).optional(),To=(e={},t={})=>{const n=Array.isArray(e)?[...e]:{...e};return Object.entries(t||{}).forEach(([e,t])=>{t&&"object"==typeof t&&!Array.isArray(t)?n[e]=To(n[e]||{},t):n[e]=t}),n},Bo=e=>e&&"object"==typeof e?JSON.parse(JSON.stringify(e)):null,$o=(e,t)=>{const n=Bo(e)||{},r=Bo(t)||{};return To(n,r)},Oo=(e=h,{defaultConfig:t}={})=>{const n=Bo(t)||{},o=Array.isArray(n.generators)?n.generators:[],{server:a,generators:i,...s}=n,l=e=>To(JSON.parse(JSON.stringify(s)),e||{}),c=(e,t)=>{if(o.length>0){const n=o.filter(t=>t?.type===e);if(n.length>0&&t){const e=n.find(e=>e.model?.repo_id===t);if(e)return l(e)}}return Object.keys(s).length>0?l({}):null};return _o.router({getCapabilities:_o.procedure.input(Mo).query(async({input:t})=>{console.log("[Server] Get Capabilities:",t);const n=t||{type:"ggml-llm"},{type:r="ggml-llm",config:o,currentClientCapabilities:a=null,options:i={}}=n,s=Bo(o),l=e.getModelIdentifier(r,s),u=c(r,l),d=$o(u,o);if(0===Object.keys(d).length)throw new Error("Buttress server missing generator configuration");return d.backend=d.backend||{},d.backend.type||(d.backend.type=r),e.getCapabilities(r,a,{...i,config:d})}),startGenerator:_o.procedure.input(Eo).mutation(async({input:t})=>{console.log("[Server] Start Generator:",t);const{type:n,config:r}=t,o=Bo(r),a=e.getModelIdentifier(n,o),i=c(n,a),s=$o(i,r);if(0===Object.keys(s).length)throw new Error("Buttress server missing generator configuration");return s.backend=s.backend||{},s.backend.type||(s.backend.type=n),e.startGenerator(n,s)}),finalizeGenerator:_o.procedure.input(So).mutation(async({input:t})=>(console.log("[Server] Finalize Generator:",t),e.finalizeGenerator(t.id))),ggmlLlm:_o.router({initContext:_o.procedure.input(ko).subscription(({input:t})=>te(n=>{console.log("[Server] Init Context:",t);let r=!0;return(async()=>{try{const o={...t.property,onProgress:e=>{r&&n.next({progress:e})}};await new Promise(e=>setTimeout(e));const a=await e.ggmlLlm.initContext(t.id,o);if(r){const{download:e,...t}=a||{};n.next({result:t}),n.complete()}}catch(e){r&&n.error(e)}})(),()=>{r=!1}})),createCompletion:_o.procedure.input(ko).mutation(async({input:t})=>{console.log("[Server] createCompletion:",t);const n=`completion-${Date.now()}-${Math.random().toString(36).slice(2,9)}`;try{const r=await e.ggmlLlm.completion(t.id,t.property);if(!r)throw new Error("Failed to create completion stream");return Co.set(n,{stream:r,cancel:null,generatorId:t.id}),{completionId:n}}catch(e){throw console.error("[Server] createCompletion error:",e),e}}),completion:_o.procedure.input(ko).subscription(({input:e})=>te(t=>{const n=e.property;console.log("[Server] completion subscription:",n);const r=Co.get(n);if(!r)return t.error(new Error(`Unknown completion ID: ${n}`)),()=>{};const{stream:o}=r,a=((e,t)=>{const n=e.getReader();let r=!0;return(async()=>{try{for(;r;){const{value:e,done:o}=await n.read();if(o||!r)break;if(e){const n=wo.decode(e,{stream:!0});t.next(n)}}t.complete()}catch(e){t.error(e)}finally{n.cancel().catch(()=>{})}})(),()=>{r=!1,n.cancel().catch(()=>{})}})(o,t);return r.cancel=a,()=>{console.log("[Server] completion subscription closed:",n),a(),Co.delete(n)}})),cancelCompletion:_o.procedure.input(ko).mutation(async({input:e})=>{const t=e.property;console.log("[Server] cancelCompletion:",t);const n=Co.get(t);return n?(n.cancel&&n.cancel(),Co.delete(t),{success:!0}):{success:!1,error:"Unknown completion ID"}}),tokenize:_o.procedure.input(ko).mutation(async({input:t})=>(console.log("[Server] Tokenize:",t),e.ggmlLlm.tokenize(t.id,t.property))),detokenize:_o.procedure.input(ko).mutation(async({input:t})=>(console.log("[Server] Detokenize:",t),e.ggmlLlm.detokenize(t.id,t.property))),applyChatTemplate:_o.procedure.input(ko).mutation(async({input:t})=>(console.log("[Server] Apply Chat Template:",t),e.ggmlLlm.applyChatTemplate(t.id,t.property))),releaseContext:_o.procedure.input(ko).mutation(async({input:t})=>(console.log("[Server] Release Context:",t),e.ggmlLlm.releaseContext(t.id,t.property)))}),ggmlStt:_o.router({initContext:_o.procedure.input(ko).subscription(({input:t})=>te(n=>{console.log("[Server] Init STT Context:",t);let r=!0;return(async()=>{try{const o={...t.property,onProgress:e=>{r&&n.next({progress:e})}};await new Promise(e=>setTimeout(e));const a=await e.ggmlStt.initContext(t.id,o);if(r){const{download:e,...t}=a||{};n.next({result:t}),n.complete()}}catch(e){r&&n.error(e)}})(),()=>{r=!1}})),transcribeData:_o.procedure.input(r.instanceof(FormData)).mutation(async({input:t})=>{const n=t,r=n.get("id"),o=n.get("options"),a=n.get("audioData"),i=o?JSON.parse(o):{};let s;if(a instanceof Blob||a&&"function"==typeof a.arrayBuffer){const e=await a.arrayBuffer();s=Buffer.from(e)}else s=Buffer.isBuffer(a)?a:a instanceof ArrayBuffer?Buffer.from(a):a;return console.log("[Server] Transcribe Data:",{id:r,audioDataLength:s?.length||0,options:i}),e.ggmlStt.transcribeData(r,{audioData:s,options:i})}),releaseContext:_o.procedure.input(ko).mutation(async({input:t})=>(console.log("[Server] Release STT Context:",t),e.ggmlStt.releaseContext(t.id,t.property)))}),status:mo(e)})},No=({backend:e=h,router:t,createContext:n=()=>({}),defaultConfig:r,maxBodySize:o=104857600,enableOpenAICompat:a="1"===process.env.ENABLE_OPENAI_COMPAT_ENDPOINT}={})=>{const i=t||Oo(e,{defaultConfig:r}),s=a?bo({backend:e,defaultConfig:r}):null;return{server:It({router:i,createContext:n,basePath:"/trpc/",maxBodySize:o,middleware:async(e,t,n)=>{const r=e.headers["content-length"],o=e.headers["content-type"];console.log(`[Request] ${e.method} ${e.url} content-length=${r} content-type=${o}`);const a=new URL(e.url,`http://${e.headers.host}`);if("/status"!==a.pathname&&"/status/"!==a.pathname){if(s&&a.pathname.startsWith("/oai-compat/"))try{if(await s(e,t,a))return}catch(e){return console.error("[OpenAI] Handler error:",e),t.writeHead(500,{"Content-Type":"application/json"}),void t.end(JSON.stringify({error:{message:e.message,type:"server_error"}}))}n()}else(async e=>{try{const t=g.join(vo,"..","public","status.html"),n=await(0,y.readFile)(t,"utf-8");return e.writeHead(200,{"Content-Type":"text/html; charset=utf-8"}),e.end(n),!0}catch(t){return e.writeHead(404,{"Content-Type":"text/plain"}),e.end("Status page not found"),!0}})(t)}}),router:i,backend:e,openaiEnabled:a}},Io=async({port:e=2080,backend:t,router:n,createContext:r,defaultConfig:o,maxBodySize:a,enableOpenAICompat:i}={})=>{const{server:s,router:l,backend:c,openaiEnabled:u}=No({backend:t,router:n,createContext:r,defaultConfig:o,maxBodySize:a,enableOpenAICompat:i});return s.server&&(void 0!==s.server.headersTimeout&&(s.server.headersTimeout=0),void 0!==s.server.requestTimeout&&(s.server.requestTimeout=0),void 0!==s.server.timeout&&(s.server.timeout=0)),await new Promise(t=>s.listen(e,t)),console.log(`[Server] Status page available at http://localhost:${e}/status`),{server:s,router:l,port:e,backend:c,openaiEnabled:u}},{testGgmlLlmCapabilities:Po,testGgmlSttCapabilities:qo,showModelsTable:Ao,showSttModelsTable:zo}=h;export{uo as checkAndNotifyUpdates,lo as checkForUpdates,co as compareVersions,Oo as createRouter,No as createServer,io as currentVersion,Ao as showModelsTable,zo as showSttModelsTable,Io as startServer,Po as testGgmlLlmCapabilities,qo as testGgmlSttCapabilities};
1
+ import{createRequire as e}from"node:module";const t=e(import.meta.url);import{initTRPC as n}from"@trpc/server";import{z as o}from"zod";import{gguf as r}from"@huggingface/gguf";import{getBackendDevicesInfo as a,isLibVariantAvailable as i,loadModel as s}from"@fugood/llama.node";import{initWhisper as l}from"@fugood/whisper.node";import{default as c}from"@iarna/toml";var u,d,p={285:e=>{e.exports=function(e,t){if("string"==typeof e)return i(e);if("number"==typeof e)return a(e,t);return null},e.exports.format=a,e.exports.parse=i;var t=/\B(?=(\d{3})+(?!\d))/g,n=/(?:\.0*|(\.[^0]+)0+)$/,o={b:1,kb:1024,mb:1<<20,gb:1<<30,tb:Math.pow(1024,4),pb:Math.pow(1024,5)},r=/^((-|\+)?(\d+(?:\.\d+)?)) *(kb|mb|gb|tb|pb)$/i;function a(e,r){if(!Number.isFinite(e))return null;var a=Math.abs(e),i=r&&r.thousandsSeparator||"",s=r&&r.unitSeparator||"",l=r&&void 0!==r.decimalPlaces?r.decimalPlaces:2,c=Boolean(r&&r.fixedDecimals),u=r&&r.unit||"";u&&o[u.toLowerCase()]||(u=a>=o.pb?"PB":a>=o.tb?"TB":a>=o.gb?"GB":a>=o.mb?"MB":a>=o.kb?"KB":"B");var d=(e/o[u.toLowerCase()]).toFixed(l);return c||(d=d.replace(n,"$1")),i&&(d=d.split(".").map(function(e,n){return 0===n?e.replace(t,i):e}).join(".")),d+s+u}function i(e){if("number"==typeof e&&!isNaN(e))return e;if("string"!=typeof e)return null;var t,n=r.exec(e),a="b";return n?(t=parseFloat(n[1]),a=n[4].toLowerCase()):(t=parseInt(e,10),a="b"),isNaN(t)?null:Math.floor(o[a]*t)}}},m={};function f(e){var t=m[e];if(void 0!==t)return t.exports;var n=m[e]={exports:{}};return p[e](n,n.exports,f),n.exports}d=Object.getPrototypeOf?e=>Object.getPrototypeOf(e):e=>e.__proto__,f.t=function(e,t){if(1&t&&(e=this(e)),8&t)return e;if("object"==typeof e&&e){if(4&t&&e.__esModule)return e;if(16&t&&"function"==typeof e.then)return e}var n=Object.create(null);f.r(n);var o={};u=u||[null,d({}),d([]),d(d)];for(var r=2&t&&e;("object"==typeof r||"function"==typeof r)&&!~u.indexOf(r);r=d(r))Object.getOwnPropertyNames(r).forEach(t=>o[t]=()=>e[t]);return o.default=()=>e,f.d(n,o),n},f.d=(e,t)=>{for(var n in t)f.o(t,n)&&!f.o(e,n)&&Object.defineProperty(e,n,{enumerable:!0,get:t[n]})},f.o=(e,t)=>Object.prototype.hasOwnProperty.call(e,t),f.r=e=>{"undefined"!=typeof Symbol&&Symbol.toStringTag&&Object.defineProperty(e,Symbol.toStringTag,{value:"Module"}),Object.defineProperty(e,"__esModule",{value:!0})};var h={};f.r(h),f.d(h,{finalizeGenerator:()=>ur,generatorRegistry:()=>ir,getCapabilities:()=>Qo,getModelIdentifier:()=>mr,ggmlLlm:()=>dr,ggmlStt:()=>pr,globalDownloadManager:()=>sr,showModelsTable:()=>nr,showSttModelsTable:()=>rr,startGenerator:()=>cr,startModelDownload:()=>hr,status:()=>fr,testGgmlLlmCapabilities:()=>or,testGgmlSttCapabilities:()=>ar});const y=t("node:fs/promises"),g=t("node:path"),b=t("node:url"),w={PARSE_ERROR:-32700,BAD_REQUEST:-32600,INTERNAL_SERVER_ERROR:-32603,NOT_IMPLEMENTED:-32603,BAD_GATEWAY:-32603,SERVICE_UNAVAILABLE:-32603,GATEWAY_TIMEOUT:-32603,UNAUTHORIZED:-32001,PAYMENT_REQUIRED:-32002,FORBIDDEN:-32003,NOT_FOUND:-32004,METHOD_NOT_SUPPORTED:-32005,TIMEOUT:-32008,CONFLICT:-32009,PRECONDITION_FAILED:-32012,PAYLOAD_TOO_LARGE:-32013,UNSUPPORTED_MEDIA_TYPE:-32015,UNPROCESSABLE_CONTENT:-32022,PRECONDITION_REQUIRED:-32028,TOO_MANY_REQUESTS:-32029,CLIENT_CLOSED_REQUEST:-32099},v={[-32700]:"PARSE_ERROR",[-32600]:"BAD_REQUEST",[-32603]:"INTERNAL_SERVER_ERROR",[-32001]:"UNAUTHORIZED",[-32002]:"PAYMENT_REQUIRED",[-32003]:"FORBIDDEN",[-32004]:"NOT_FOUND",[-32005]:"METHOD_NOT_SUPPORTED",[-32008]:"TIMEOUT",[-32009]:"CONFLICT",[-32012]:"PRECONDITION_FAILED",[-32013]:"PAYLOAD_TOO_LARGE",[-32015]:"UNSUPPORTED_MEDIA_TYPE",[-32022]:"UNPROCESSABLE_CONTENT",[-32028]:"PRECONDITION_REQUIRED",[-32029]:"TOO_MANY_REQUESTS",[-32099]:"CLIENT_CLOSED_REQUEST"};function _(e){return!!e&&!Array.isArray(e)&&"object"==typeof e}const x="function"==typeof Symbol&&!!Symbol.asyncIterator;function S(e){return x&&_(e)&&Symbol.asyncIterator in e}const C=e=>e();function k(e){return e}var T=Object.create,M=Object.defineProperty,E=Object.getOwnPropertyDescriptor,$=Object.getOwnPropertyNames,B=Object.getPrototypeOf,I=Object.prototype.hasOwnProperty,O=(e,t)=>function(){return t||(0,e[$(e)[0]])((t={exports:{}}).exports,t),t.exports},N=(e,t,n)=>(n=null!=e?T(B(e)):{},((e,t,n,o)=>{if(t&&"object"==typeof t||"function"==typeof t)for(var r,a=$(t),i=0,s=a.length;i<s;i++)r=a[i],I.call(e,r)||r===n||M(e,r,{get:(e=>t[e]).bind(null,r),enumerable:!(o=E(t,r))||o.enumerable});return e})(!t&&e&&e.__esModule?n:M(n,"default",{value:e,enumerable:!0}),e));const P={PARSE_ERROR:400,BAD_REQUEST:400,UNAUTHORIZED:401,PAYMENT_REQUIRED:402,FORBIDDEN:403,NOT_FOUND:404,METHOD_NOT_SUPPORTED:405,TIMEOUT:408,CONFLICT:409,PRECONDITION_FAILED:412,PAYLOAD_TOO_LARGE:413,UNSUPPORTED_MEDIA_TYPE:415,UNPROCESSABLE_CONTENT:422,PRECONDITION_REQUIRED:428,TOO_MANY_REQUESTS:429,CLIENT_CLOSED_REQUEST:499,INTERNAL_SERVER_ERROR:500,NOT_IMPLEMENTED:501,BAD_GATEWAY:502,SERVICE_UNAVAILABLE:503,GATEWAY_TIMEOUT:504};function q(e){var t;return null!==(t=P[e])&&void 0!==t?t:500}function z(e){return q(e.code)}var D=O({"../../node_modules/.pnpm/@oxc-project+runtime@0.72.2/node_modules/@oxc-project/runtime/src/helpers/typeof.js"(e,t){function n(e){return t.exports=n="function"==typeof Symbol&&"symbol"==typeof Symbol.iterator?function(e){return typeof e}:function(e){return e&&"function"==typeof Symbol&&e.constructor===Symbol&&e!==Symbol.prototype?"symbol":typeof e},t.exports.__esModule=!0,t.exports.default=t.exports,n(e)}t.exports=n,t.exports.__esModule=!0,t.exports.default=t.exports}}),A=O({"../../node_modules/.pnpm/@oxc-project+runtime@0.72.2/node_modules/@oxc-project/runtime/src/helpers/toPrimitive.js"(e,t){var n=D().default;t.exports=function(e,t){if("object"!=n(e)||!e)return e;var o=e[Symbol.toPrimitive];if(void 0!==o){var r=o.call(e,t||"default");if("object"!=n(r))return r;throw new TypeError("@@toPrimitive must return a primitive value.")}return("string"===t?String:Number)(e)},t.exports.__esModule=!0,t.exports.default=t.exports}}),F=O({"../../node_modules/.pnpm/@oxc-project+runtime@0.72.2/node_modules/@oxc-project/runtime/src/helpers/toPropertyKey.js"(e,t){var n=D().default,o=A();t.exports=function(e){var t=o(e,"string");return"symbol"==n(t)?t:t+""},t.exports.__esModule=!0,t.exports.default=t.exports}}),R=O({"../../node_modules/.pnpm/@oxc-project+runtime@0.72.2/node_modules/@oxc-project/runtime/src/helpers/defineProperty.js"(e,t){var n=F();t.exports=function(e,t,o){return(t=n(t))in e?Object.defineProperty(e,t,{value:o,enumerable:!0,configurable:!0,writable:!0}):e[t]=o,e},t.exports.__esModule=!0,t.exports.default=t.exports}}),U=O({"../../node_modules/.pnpm/@oxc-project+runtime@0.72.2/node_modules/@oxc-project/runtime/src/helpers/objectSpread2.js"(e,t){var n=R();function o(e,t){var n=Object.keys(e);if(Object.getOwnPropertySymbols){var o=Object.getOwnPropertySymbols(e);t&&(o=o.filter(function(t){return Object.getOwnPropertyDescriptor(e,t).enumerable})),n.push.apply(n,o)}return n}t.exports=function(e){for(var t=1;t<arguments.length;t++){var r=null!=arguments[t]?arguments[t]:{};t%2?o(Object(r),!0).forEach(function(t){n(e,t,r[t])}):Object.getOwnPropertyDescriptors?Object.defineProperties(e,Object.getOwnPropertyDescriptors(r)):o(Object(r)).forEach(function(t){Object.defineProperty(e,t,Object.getOwnPropertyDescriptor(r,t))})}return e},t.exports.__esModule=!0,t.exports.default=t.exports}}),j=N(U(),1);function L(e){const{path:t,error:n,config:o}=e,{code:r}=e.error,a={message:n.message,code:w[r],data:{code:r,httpStatus:z(n)}};return o.isDev&&"string"==typeof e.error.stack&&(a.data.stack=e.error.stack),"string"==typeof t&&(a.data.path=t),o.errorFormatter((0,j.default)((0,j.default)({},e),{},{shape:a}))}var G=N(R(),1),H=class extends Error{};function J(e){if(e instanceof V)return e;if(e instanceof Error&&"TRPCError"===e.name)return e;const t=new V({code:"INTERNAL_SERVER_ERROR",cause:e});return e instanceof Error&&e.stack&&(t.stack=e.stack),t}var V=class extends Error{constructor(e){var t,n,o;const r=function(e){if(e instanceof Error)return e;const t=typeof e;return"undefined"!==t&&"function"!==t&&null!==e?"object"!==t?new Error(String(e)):_(e)?Object.assign(new H,e):void 0:void 0}(e.cause);super(null!==(t=null!==(n=e.message)&&void 0!==n?n:null==r?void 0:r.message)&&void 0!==t?t:e.code,{cause:r}),(0,G.default)(this,"cause",void 0),(0,G.default)(this,"code",void 0),this.code=e.code,this.name="TRPCError",null!==(o=this.cause)&&void 0!==o||(this.cause=r)}},K=N(U(),1);function Y(e,t){return"error"in t?(0,K.default)((0,K.default)({},t),{},{error:e.transformer.output.serialize(t.error)}):"data"in t.result?(0,K.default)((0,K.default)({},t),{},{result:(0,K.default)((0,K.default)({},t.result),{},{data:e.transformer.output.serialize(t.result.data)})}):t}function W(e,t){return Array.isArray(t)?t.map(t=>Y(e,t)):Y(e,t)}Error;N(U(),1);async function Q(e,t){const{_def:n}=e;let o=n.procedures[t];for(;!o;){const e=Object.keys(n.lazy).find(e=>t.startsWith(e));if(!e)return null;const r=n.lazy[e];await r.load(),o=n.procedures[t]}return o}const Z=Symbol();function X(e){return Array.isArray(e)&&e[2]===Z}function ee(e){return"object"==typeof e&&null!==e&&"subscribe"in e}function te(e){const t={subscribe(t){let n=null,o=!1,r=!1,a=!1;function i(){null!==n?r||(r=!0,"function"==typeof n?n():n&&n.unsubscribe()):a=!0}return n=e({next(e){var n;o||null===(n=t.next)||void 0===n||n.call(t,e)},error(e){var n;o||(o=!0,null===(n=t.error)||void 0===n||n.call(t,e),i())},complete(){var e;o||(o=!0,null===(e=t.complete)||void 0===e||e.call(t),i())}}),a&&i(),{unsubscribe:i}},pipe:(...e)=>e.reduce(ne,t)};return t}function ne(e,t){return t(e)}function oe(e,t){const n=function(e,t){let n=null;const o=()=>{null==n||n.unsubscribe(),n=null,t.removeEventListener("abort",o)};return new ReadableStream({start(r){n=e.subscribe({next(e){r.enqueue({ok:!0,value:e})},error(e){r.enqueue({ok:!1,error:e}),r.close()},complete(){r.close()}}),t.aborted?o():t.addEventListener("abort",o,{once:!0})},cancel(){o()}})}(e,t),o=n.getReader(),r={async next(){const e=await o.read();if(e.done)return{value:void 0,done:!0};const{value:t}=e;if(!t.ok)throw t.error;return{value:t.value,done:!1}},return:async()=>(await o.cancel(),{value:void 0,done:!0})};return{[Symbol.asyncIterator]:()=>r}}function re(e){let t;try{t=JSON.parse(e)}catch(e){throw new V({code:"PARSE_ERROR",message:"Not JSON-parsable query params",cause:e})}return function(e){try{if(null===e)return null;if(!_(e))throw new Error("Expected object");const t=Object.entries(e).filter(([e,t])=>"string"!=typeof t);if(t.length>0)throw new Error(`Expected connectionParams to be string values. Got ${t.map(([e,t])=>`${e}: ${typeof t}`).join(", ")}`);return e}catch(e){throw new V({code:"PARSE_ERROR",message:"Invalid connection params shape",cause:e})}}(t)}var ae=N(U(),1);function ie(e){let t=null;const n=Symbol.for("@trpc/server/http/memo");let o=n;return{read:async()=>{var r;return o!==n||(null!==(r=t)&&void 0!==r||(t=e().catch(e=>{if(e instanceof V)throw e;throw new V({code:"BAD_REQUEST",message:e instanceof Error?e.message:"Invalid input",cause:e})})),o=await t,t=null),o},result:()=>o!==n?o:void 0}}const se={isMatch(e){var t;return!!(null===(t=e.headers.get("content-type"))||void 0===t?void 0:t.startsWith("application/json"))},async parse(e){var t;const{req:n}=e,o="1"===e.searchParams.get("batch"),r=o?e.path.split(","):[e.path],a=ie(async()=>{let t;if("GET"===n.method){const n=e.searchParams.get("input");n&&(t=JSON.parse(n))}else t=await n.json();if(void 0===t)return{};if(!o)return{0:e.router._def._config.transformer.input.deserialize(t)};if(!_(t))throw new V({code:"BAD_REQUEST",message:'"input" needs to be an object when doing a batch call'});const a={};for(const n of r.keys()){const o=t[n];void 0!==o&&(a[n]=e.router._def._config.transformer.input.deserialize(o))}return a}),i=await Promise.all(r.map(async(t,n)=>{const o=await Q(e.router,t);return{path:t,procedure:o,getRawInput:async()=>{let t=(await a.read())[n];if("subscription"===(null==o?void 0:o._def.type)){var r,i;const n=null!==(r=null!==(i=e.headers.get("last-event-id"))&&void 0!==i?i:e.searchParams.get("lastEventId"))&&void 0!==r?r:e.searchParams.get("Last-Event-Id");if(n)if(_(t))t=(0,ae.default)((0,ae.default)({},t),{},{lastEventId:n});else{var s;null!==(s=t)&&void 0!==s||(t={lastEventId:n})}}return t},result:()=>{var e;return null===(e=a.result())||void 0===e?void 0:e[n]}}})),s=new Set(i.map(e=>{var t;return null===(t=e.procedure)||void 0===t?void 0:t._def.type}).filter(Boolean));if(s.size>1)throw new V({code:"BAD_REQUEST",message:`Cannot mix procedure types in call: ${Array.from(s).join(", ")}`});const l=null!==(t=s.values().next().value)&&void 0!==t?t:"unknown",c=e.searchParams.get("connectionParams");return{isBatchCall:o,accept:n.headers.get("trpc-accept"),calls:i,type:l,connectionParams:null===c?null:re(c),signal:n.signal,url:e.url}}},le=[se,{isMatch(e){var t;return!!(null===(t=e.headers.get("content-type"))||void 0===t?void 0:t.startsWith("multipart/form-data"))},async parse(e){const{req:t}=e;if("POST"!==t.method)throw new V({code:"METHOD_NOT_SUPPORTED",message:"Only POST requests are supported for multipart/form-data requests"});const n=ie(async()=>await t.formData()),o=await Q(e.router,e.path);return{accept:null,calls:[{path:e.path,getRawInput:n.read,result:n.result,procedure:o}],isBatchCall:!1,type:"mutation",connectionParams:null,signal:t.signal,url:e.url}}},{isMatch(e){var t;return!!(null===(t=e.headers.get("content-type"))||void 0===t?void 0:t.startsWith("application/octet-stream"))},async parse(e){const{req:t}=e;if("POST"!==t.method)throw new V({code:"METHOD_NOT_SUPPORTED",message:"Only POST requests are supported for application/octet-stream requests"});const n=ie(async()=>t.body);return{calls:[{path:e.path,getRawInput:n.read,result:n.result,procedure:await Q(e.router,e.path)}],isBatchCall:!1,accept:null,type:"mutation",connectionParams:null,signal:t.signal,url:e.url}}}];async function ce(e){const t=function(e){const t=le.find(t=>t.isMatch(e));if(t)return t;if(!t&&"GET"===e.method)return se;throw new V({code:"UNSUPPORTED_MEDIA_TYPE",message:e.headers.has("content-type")?`Unsupported content-type "${e.headers.get("content-type")}`:"Missing content-type header"})}(e.req);return await t.parse(e)}function ue(e){return _(e)&&"AbortError"===e.name}function de(e="AbortError"){throw new DOMException(e,"AbortError")}function pe(e){return"[object Object]"===Object.prototype.toString.call(e)}var me=N(R(),1);let fe;const he=new WeakMap,ye=()=>{};fe=Symbol.toStringTag;var ge,be,we,ve,_e=class e{constructor(e){(0,me.default)(this,"promise",void 0),(0,me.default)(this,"subscribers",[]),(0,me.default)(this,"settlement",null),(0,me.default)(this,fe,"Unpromise"),this.promise="function"==typeof e?new Promise(e):e;const t=this.promise.then(e=>{const{subscribers:t}=this;this.subscribers=null,this.settlement={status:"fulfilled",value:e},null==t||t.forEach(({resolve:t})=>{t(e)})});"catch"in t&&t.catch(e=>{const{subscribers:t}=this;this.subscribers=null,this.settlement={status:"rejected",reason:e},null==t||t.forEach(({reject:t})=>{t(e)})})}subscribe(){let e,t;const{settlement:n}=this;if(null===n){if(null===this.subscribers)throw new Error("Unpromise settled but still has subscribers");const n=function(){let e,t;const n=new Promise((n,o)=>{e=n,t=o});return{promise:n,resolve:e,reject:t}}();this.subscribers=(o=this.subscribers,r=n,[...o,r]),e=n.promise,t=()=>{null!==this.subscribers&&(this.subscribers=function(e,t){const n=e.indexOf(t);return-1!==n?function(e,t){return[...e.slice(0,t),...e.slice(t+1)]}(e,n):e}(this.subscribers,n))}}else{const{status:o}=n;e="fulfilled"===o?Promise.resolve(n.value):Promise.reject(n.reason),t=ye}var o,r;return Object.assign(e,{unsubscribe:t})}then(e,t){const n=this.subscribe(),{unsubscribe:o}=n;return Object.assign(n.then(e,t),{unsubscribe:o})}catch(e){const t=this.subscribe(),{unsubscribe:n}=t;return Object.assign(t.catch(e),{unsubscribe:n})}finally(e){const t=this.subscribe(),{unsubscribe:n}=t;return Object.assign(t.finally(e),{unsubscribe:n})}static proxy(t){const n=e.getSubscribablePromise(t);return void 0!==n?n:e.createSubscribablePromise(t)}static createSubscribablePromise(t){const n=new e(t);return he.set(t,n),he.set(n,n),n}static getSubscribablePromise(e){return he.get(e)}static resolve(t){const n="object"==typeof t&&null!==t&&"then"in t&&"function"==typeof t.then?t:Promise.resolve(t);return e.proxy(n).subscribe()}static async any(t){const n=(Array.isArray(t)?t:[...t]).map(e.resolve);try{return await Promise.any(n)}finally{n.forEach(({unsubscribe:e})=>{e()})}}static async race(t){const n=(Array.isArray(t)?t:[...t]).map(e.resolve);try{return await Promise.race(n)}finally{n.forEach(({unsubscribe:e})=>{e()})}}static async raceReferences(e){const t=e.map(xe);try{return await Promise.race(t)}finally{for(const e of t)e.unsubscribe()}}};function xe(e){return _e.proxy(e).then(()=>[e])}function Se(e,t){const n=e,o=n[Symbol.dispose];return n[Symbol.dispose]=()=>{t(),null==o||o()},n}function Ce(e,t){const n=e,o=n[Symbol.asyncDispose];return n[Symbol.asyncDispose]=async()=>{await t(),await(null==o?void 0:o())},n}null!==(be=(ge=Symbol).dispose)&&void 0!==be||(ge.dispose=Symbol()),null!==(ve=(we=Symbol).asyncDispose)&&void 0!==ve||(we.asyncDispose=Symbol());const ke=Symbol();function Te(e){let t=null;return Se({start(){if(t)throw new Error("Timer already started");return new Promise(n=>{t=setTimeout(()=>n(ke),e)})}},()=>{t&&clearTimeout(t)})}var Me=O({"../../node_modules/.pnpm/@oxc-project+runtime@0.72.2/node_modules/@oxc-project/runtime/src/helpers/usingCtx.js"(e,t){t.exports=function(){var e="function"==typeof SuppressedError?SuppressedError:function(e,t){var n=Error();return n.name="SuppressedError",n.error=e,n.suppressed=t,n},t={},n=[];function o(e,t){if(null!=t){if(Object(t)!==t)throw new TypeError("using declarations can only be used with objects, functions, null, or undefined.");if(e)var o=t[Symbol.asyncDispose||Symbol.for("Symbol.asyncDispose")];if(void 0===o&&(o=t[Symbol.dispose||Symbol.for("Symbol.dispose")],e))var r=o;if("function"!=typeof o)throw new TypeError("Object is not disposable.");r&&(o=function(){try{r.call(t)}catch(e){return Promise.reject(e)}}),n.push({v:t,d:o,a:e})}else e&&n.push({d:t,a:e});return t}return{e:t,u:o.bind(null,!1),a:o.bind(null,!0),d:function(){var o,r=this.e,a=0;function i(){for(;o=n.pop();)try{if(!o.a&&1===a)return a=0,n.push(o),Promise.resolve().then(i);if(o.d){var e=o.d.call(o.v);if(o.a)return a|=2,Promise.resolve(e).then(i,s)}else a|=1}catch(e){return s(e)}if(1===a)return r!==t?Promise.reject(r):Promise.resolve();if(r!==t)throw r}function s(n){return r=r!==t?new e(n,r):n,i()}return i()}}},t.exports.__esModule=!0,t.exports.default=t.exports}}),Ee=O({"../../node_modules/.pnpm/@oxc-project+runtime@0.72.2/node_modules/@oxc-project/runtime/src/helpers/OverloadYield.js"(e,t){t.exports=function(e,t){this.v=e,this.k=t},t.exports.__esModule=!0,t.exports.default=t.exports}}),$e=O({"../../node_modules/.pnpm/@oxc-project+runtime@0.72.2/node_modules/@oxc-project/runtime/src/helpers/awaitAsyncGenerator.js"(e,t){var n=Ee();t.exports=function(e){return new n(e,0)},t.exports.__esModule=!0,t.exports.default=t.exports}}),Be=O({"../../node_modules/.pnpm/@oxc-project+runtime@0.72.2/node_modules/@oxc-project/runtime/src/helpers/wrapAsyncGenerator.js"(e,t){var n=Ee();function o(e){var t,o;function r(t,o){try{var i=e[t](o),s=i.value,l=s instanceof n;Promise.resolve(l?s.v:s).then(function(n){if(l){var o="return"===t?"return":"next";if(!s.k||n.done)return r(o,n);n=e[o](n).value}a(i.done?"return":"normal",n)},function(e){r("throw",e)})}catch(e){a("throw",e)}}function a(e,n){switch(e){case"return":t.resolve({value:n,done:!0});break;case"throw":t.reject(n);break;default:t.resolve({value:n,done:!1})}(t=t.next)?r(t.key,t.arg):o=null}this._invoke=function(e,n){return new Promise(function(a,i){var s={key:e,arg:n,resolve:a,reject:i,next:null};o?o=o.next=s:(t=o=s,r(e,n))})},"function"!=typeof e.return&&(this.return=void 0)}o.prototype["function"==typeof Symbol&&Symbol.asyncIterator||"@@asyncIterator"]=function(){return this},o.prototype.next=function(e){return this._invoke("next",e)},o.prototype.throw=function(e){return this._invoke("throw",e)},o.prototype.return=function(e){return this._invoke("return",e)},t.exports=function(e){return function(){return new o(e.apply(this,arguments))}},t.exports.__esModule=!0,t.exports.default=t.exports}}),Ie=N(Me(),1),Oe=N($e(),1),Ne=N(Be(),1);function Pe(e){const t=e[Symbol.asyncIterator]();return t[Symbol.asyncDispose]?t:Ce(t,async()=>{var e;await(null===(e=t.return)||void 0===e?void 0:e.call(t))})}function qe(){return(qe=(0,Ne.default)(function*(e,t){try{var n=(0,Ie.default)();const o=n.a(Pe(e)),r=n.u(Te(t.maxDurationMs)).start();let a;for(;;){if(a=yield(0,Oe.default)(_e.race([o.next(),r])),a===ke&&de(),a.done)return a;yield a.value,a=null}}catch(e){n.e=e}finally{yield(0,Oe.default)(n.d())}})).apply(this,arguments)}function ze(){return(ze=(0,Ne.default)(function*(e,t){try{var n=(0,Ie.default)();const o=n.a(Pe(e));let r;const a=n.u(Te(t.gracePeriodMs));let i=t.count,s=new Promise(()=>{});for(;;){if(r=yield(0,Oe.default)(_e.race([o.next(),s])),r===ke&&de(),r.done)return r.value;yield r.value,0===--i&&(s=a.start()),r=null}}catch(e){n.e=e}finally{yield(0,Oe.default)(n.d())}})).apply(this,arguments)}function De(){let e,t;return{promise:new Promise((n,o)=>{e=n,t=o}),resolve:e,reject:t}}var Ae=N(Me(),1),Fe=N($e(),1),Re=N(Be(),1);function Ue(){let e="idle",t=De();const n=[],o=new Set,r=[];function a(n){if("pending"!==e)return;const a=function(e,t){const n=e[Symbol.asyncIterator]();let o="idle";function r(){o="done",t=()=>{}}return{pull:function(){"idle"===o&&(o="pending",n.next().then(e=>{if(e.done)return o="done",t({status:"return",value:e.value}),void r();o="idle",t({status:"yield",value:e.value})}).catch(e=>{t({status:"error",error:e}),r()}))},destroy:async()=>{var e;r(),await(null===(e=n.return)||void 0===e?void 0:e.call(n))}}}(n,n=>{if("pending"===e){switch(n.status){case"yield":r.push([a,n]);break;case"return":o.delete(a);break;case"error":r.push([a,n]),o.delete(a)}t.resolve()}});o.add(a),a.pull()}return{add(t){switch(e){case"idle":n.push(t);break;case"pending":a(t)}},[Symbol.asyncIterator]:()=>(0,Re.default)(function*(){try{var i=(0,Ae.default)();if("idle"!==e)throw new Error("Cannot iterate twice");e="pending";i.a(Ce({},async()=>{e="done";const n=[];if(await Promise.all(Array.from(o.values()).map(async e=>{try{await e.destroy()}catch(e){n.push(e)}})),r.length=0,o.clear(),t.resolve(),n.length>0)throw new AggregateError(n)}));for(;n.length>0;)a(n.shift());for(;o.size>0;){for(yield(0,Fe.default)(t.promise);r.length>0;){const[e,t]=r.shift();switch(t.status){case"yield":yield t.value,e.pull();break;case"error":throw t.error}}t=De()}}catch(e){i.e=e}finally{yield(0,Fe.default)(i.d())}})()}}function je(e){const t=e[Symbol.asyncIterator]();return new ReadableStream({async cancel(){var e;await(null===(e=t.return)||void 0===e?void 0:e.call(t))},async pull(e){const n=await t.next();n.done?e.close():e.enqueue(n.value)}})}var Le=N(Me(),1),Ge=N($e(),1),He=N(Be(),1);const Je=Symbol("ping");function Ve(e,t){return Ke.apply(this,arguments)}function Ke(){return(Ke=(0,He.default)(function*(e,t){try{var n=(0,Le.default)();const r=n.a(Pe(e));let a,i=r.next();for(;;)try{var o=(0,Le.default)();const e=o.u(Te(t));if(a=yield(0,Ge.default)(_e.race([i,e.start()])),a===ke){yield Je;continue}if(a.done)return a.value;i=r.next(),yield a.value,a=null}catch(e){o.e=e}finally{o.d()}}catch(e){n.e=e}finally{yield(0,Ge.default)(n.d())}})).apply(this,arguments)}var Ye=O({"../../node_modules/.pnpm/@oxc-project+runtime@0.72.2/node_modules/@oxc-project/runtime/src/helpers/asyncIterator.js"(e,t){function n(e){function t(e){if(Object(e)!==e)return Promise.reject(new TypeError(e+" is not an object."));var t=e.done;return Promise.resolve(e.value).then(function(e){return{value:e,done:t}})}return(n=function(e){this.s=e,this.n=e.next}).prototype={s:null,n:null,next:function(){return t(this.n.apply(this.s,arguments))},return:function(e){var n=this.s.return;return void 0===n?Promise.resolve({value:e,done:!0}):t(n.apply(this.s,arguments))},throw:function(e){var n=this.s.return;return void 0===n?Promise.reject(e):t(n.apply(this.s,arguments))}},new n(e)}t.exports=function(e){var t,o,r,a=2;for("undefined"!=typeof Symbol&&(o=Symbol.asyncIterator,r=Symbol.iterator);a--;){if(o&&null!=(t=e[o]))return t.call(e);if(r&&null!=(t=e[r]))return new n(t.call(e));o="@@asyncIterator",r="@@iterator"}throw new TypeError("Object is not async iterable")},t.exports.__esModule=!0,t.exports.default=t.exports}}),We=N($e(),1),Qe=N(Be(),1),Ze=N(Me(),1),Xe=N(Ye(),1);function et(e){return(_(e)||"function"==typeof e)&&"function"==typeof(null==e?void 0:e.then)&&"function"==typeof(null==e?void 0:e.catch)}var tt=class extends Error{constructor(e){super("Max depth reached at path: "+e.join(".")),this.path=e}};function nt(){return nt=(0,Qe.default)(function*(e){const{data:t}=e;let n=0;const o=Ue();function r(e){const t=n++,r=e(t);return o.add(r),t}function a(t,n){return r(function(){var o=(0,Qe.default)(function*(o){const r=s(n);r&&(t.catch(t=>{var o;null===(o=e.onError)||void 0===o||o.call(e,{error:t,path:n})}),t=Promise.reject(r));try{const e=yield(0,We.default)(t);yield[o,0,c(e,n)]}catch(t){var a,i;null===(a=e.onError)||void 0===a||a.call(e,{error:t,path:n}),yield[o,1,null===(i=e.formatError)||void 0===i?void 0:i.call(e,{error:t,path:n})]}});return function(e){return o.apply(this,arguments)}}())}function i(t,n){return r(function(){var o=(0,Qe.default)(function*(o){try{var r=(0,Ze.default)();const l=s(n);if(l)throw l;const u=r.a(Pe(t));try{for(;;){const e=yield(0,We.default)(u.next());if(e.done){yield[o,0,c(e.value,n)];break}yield[o,1,c(e.value,n)]}}catch(t){var a,i;null===(a=e.onError)||void 0===a||a.call(e,{error:t,path:n}),yield[o,2,null===(i=e.formatError)||void 0===i?void 0:i.call(e,{error:t,path:n})]}}catch(e){r.e=e}finally{yield(0,We.default)(r.d())}});return function(e){return o.apply(this,arguments)}}())}function s(t){return e.maxDepth&&t.length>e.maxDepth?new tt(t):null}function l(t,n){if(et(t))return[0,a(t,n)];if(S(t)){if(e.maxDepth&&n.length>=e.maxDepth)throw new Error("Max depth reached");return[1,i(t,n)]}return null}function c(e,t){if(void 0===e)return[[]];const n=l(e,t);if(n)return[[0],[null,...n]];if(!1===pe(o=e)||void 0!==(r=o.constructor)&&(!1===pe(a=r.prototype)||!1===a.hasOwnProperty("isPrototypeOf")))return[[e]];var o,r,a;const i={},s=[];for(const[n,o]of Object.entries(e)){const e=l(o,[...t,n]);e?(i[n]=0,s.push([n,...e])):i[n]=o}return[[i],...s]}const u={};for(const[e,n]of Object.entries(t))u[e]=c(n,[e]);yield u;let d=o;e.pingMs&&(d=Ve(o,e.pingMs));var p,m=!1,f=!1;try{for(var h,y=(0,Xe.default)(d);m=!(h=yield(0,We.default)(y.next())).done;m=!1){const e=h.value;yield e}}catch(e){f=!0,p=e}finally{try{m&&null!=y.return&&(yield(0,We.default)(y.return()))}finally{if(f)throw p}}}),nt.apply(this,arguments)}function ot(e){let t=je(function(e){return nt.apply(this,arguments)}(e));const{serialize:n}=e;return n&&(t=t.pipeThrough(new TransformStream({transform(e,t){e===Je?t.enqueue(Je):t.enqueue(n(e))}}))),t.pipeThrough(new TransformStream({transform(e,t){e===Je?t.enqueue(" "):t.enqueue(JSON.stringify(e)+"\n")}})).pipeThrough(new TextEncoderStream)}Error;var rt=O({"../../node_modules/.pnpm/@oxc-project+runtime@0.72.2/node_modules/@oxc-project/runtime/src/helpers/asyncGeneratorDelegate.js"(e,t){var n=Ee();t.exports=function(e){var t={},o=!1;function r(t,r){return o=!0,r=new Promise(function(n){n(e[t](r))}),{done:!1,value:new n(r,1)}}return t["undefined"!=typeof Symbol&&Symbol.iterator||"@@iterator"]=function(){return this},t.next=function(e){return o?(o=!1,e):r("next",e)},"function"==typeof e.throw&&(t.throw=function(e){if(o)throw o=!1,e;return r("throw",e)}),"function"==typeof e.return&&(t.return=function(e){return o?(o=!1,e):r("return",e)}),t},t.exports.__esModule=!0,t.exports.default=t.exports}}),at=N(Ye(),1),it=N($e(),1),st=N(Be(),1),lt=N(rt(),1);N(Me(),1);const ct="ping",ut="serialized-error",dt="connected",pt="return";function mt(e){var t,n,o,r,a;const{serialize:i=k}=e,s={enabled:null!==(t=null===(n=e.ping)||void 0===n?void 0:n.enabled)&&void 0!==t&&t,intervalMs:null!==(o=null===(r=e.ping)||void 0===r?void 0:r.intervalMs)&&void 0!==o?o:1e3},l=null!==(a=e.client)&&void 0!==a?a:{};if(s.enabled&&l.reconnectAfterInactivityMs&&s.intervalMs>l.reconnectAfterInactivityMs)throw new Error(`Ping interval must be less than client reconnect interval to prevent unnecessary reconnection - ping.intervalMs: ${s.intervalMs} client.reconnectAfterInactivityMs: ${l.reconnectAfterInactivityMs}`);function c(){return c=(0,st.default)(function*(){yield{event:dt,data:JSON.stringify(l)};let t,n,o=e.data;e.emitAndEndImmediately&&(o=function(e,t){return ze.apply(this,arguments)}(o,{count:1,gracePeriodMs:1})),e.maxDurationMs&&e.maxDurationMs>0&&e.maxDurationMs!==1/0&&(o=function(e,t){return qe.apply(this,arguments)}(o,{maxDurationMs:e.maxDurationMs})),s.enabled&&s.intervalMs!==1/0&&s.intervalMs>0&&(o=Ve(o,s.intervalMs));var r,a=!1,c=!1;try{for(var u,d=(0,at.default)(o);a=!(u=yield(0,it.default)(d.next())).done;a=!1)t=u.value,t!==Je?(n=X(t)?{id:t[0],data:t[1]}:{data:t},n.data=JSON.stringify(i(n.data)),yield n,t=null,n=null):yield{event:ct,data:""}}catch(e){c=!0,r=e}finally{try{a&&null!=d.return&&(yield(0,it.default)(d.return()))}finally{if(c)throw r}}}),c.apply(this,arguments)}function u(){return u=(0,st.default)(function*(){try{yield*(0,lt.default)((0,at.default)(function(){return c.apply(this,arguments)}())),yield{event:pt,data:""}}catch(o){var t,n;if(ue(o))return;const r=J(o),a=null!==(t=null===(n=e.formatError)||void 0===n?void 0:n.call(e,{error:r}))&&void 0!==t?t:null;yield{event:ut,data:JSON.stringify(i(a))}}}),u.apply(this,arguments)}const d=je(function(){return u.apply(this,arguments)}());return d.pipeThrough(new TransformStream({transform(e,t){"event"in e&&t.enqueue(`event: ${e.event}\n`),"data"in e&&t.enqueue(`data: ${e.data}\n`),"id"in e&&t.enqueue(`id: ${e.id}\n`),"comment"in e&&t.enqueue(`: ${e.comment}\n`),t.enqueue("\n\n")}})).pipeThrough(new TextEncoderStream)}const ft={"Content-Type":"text/event-stream","Cache-Control":"no-cache, no-transform","X-Accel-Buffering":"no",Connection:"keep-alive"};var ht=N(Be(),1),yt=N(U(),1);function gt(e){return C((0,ht.default)(function*(){throw e}))}const bt={mutation:["POST"],query:["GET"],subscription:["GET"]},wt={mutation:["POST"],query:["GET","POST"],subscription:["GET","POST"]};function vt(e){var t,n,o;const{ctx:r,info:a,responseMeta:i,untransformedJSON:s,errors:l=[],headers:c}=e;let u=s?function(e){const t=Array.isArray(e)?e:[e],n=new Set(t.map(e=>{var t;return"error"in e&&_(e.error.data)?"number"==typeof(null===(t=e.error.data)||void 0===t?void 0:t.httpStatus)?e.error.data.httpStatus:q(v[e.error.code]):200}));return 1!==n.size?207:n.values().next().value}(s):200;const d=!s,p=d?[]:Array.isArray(s)?s:[s],m=null!==(t=null==i?void 0:i({ctx:r,info:a,paths:null==a?void 0:a.calls.map(e=>e.path),data:p,errors:l,eagerGeneration:d,type:null!==(n=null==a||null===(o=a.calls.find(e=>{var t;return null===(t=e.procedure)||void 0===t?void 0:t._def.type}))||void 0===o||null===(o=o.procedure)||void 0===o?void 0:o._def.type)&&void 0!==n?n:"unknown"}))&&void 0!==t?t:{};if(m.headers)if(m.headers instanceof Headers)for(const[e,t]of m.headers.entries())c.append(e,t);else for(const[e,t]of Object.entries(m.headers))if(Array.isArray(t))for(const n of t)c.append(e,n);else"string"==typeof t&&c.set(e,t);return m.status&&(u=m.status),{status:u}}function _t(e){return!!_(e)&&(!!S(e)||(Object.values(e).some(et)||Object.values(e).some(S)))}async function xt(e){var t,n,o,r,a,i;const{router:s,req:l}=e,c=new Headers([["vary","trpc-accept"]]),u=s._def._config,d=new URL(l.url);if("HEAD"===l.method)return new Response(null,{status:204});const p=null===(t=null!==(n=e.allowBatching)&&void 0!==n?n:null===(o=e.batching)||void 0===o?void 0:o.enabled)||void 0===t||t,m=null!==(r=e.allowMethodOverride)&&void 0!==r&&r&&"POST"===l.method,f=await C(async()=>{try{return[void 0,await ce({req:l,path:decodeURIComponent(e.path),router:s,searchParams:d.searchParams,headers:e.req.headers,url:d})]}catch(e){return[J(e),void 0]}}),h=C(()=>{let t;return{valueOrUndefined:()=>{if(t)return t[1]},value:()=>{const[e,n]=t;if(e)throw e;return n},create:async n=>{if(t)throw new Error("This should only be called once - report a bug in tRPC");try{const o=await e.createContext({info:n});t=[void 0,o]}catch(e){t=[J(e),void 0]}}}}),y=m?wt:bt,g="application/jsonl"===l.headers.get("trpc-accept"),b=null===(a=null===(i=u.sse)||void 0===i?void 0:i.enabled)||void 0===a||a;try{const[t,n]=f;if(t)throw t;if(n.isBatchCall&&!p)throw new V({code:"BAD_REQUEST",message:"Batching is not enabled on the server"});if(g&&!n.isBatchCall)throw new V({message:"Streaming requests must be batched (you can do a batch of 1)",code:"BAD_REQUEST"});await h.create(n);const o=n.calls.map(async t=>{const o=t.procedure;try{if(e.error)throw e.error;if(!o)throw new V({code:"NOT_FOUND",message:`No procedure found on path "${t.path}"`});if(!y[o._def.type].includes(l.method))throw new V({code:"METHOD_NOT_SUPPORTED",message:`Unsupported ${l.method}-request to ${o._def.type} procedure at path "${t.path}"`});if("subscription"===o._def.type&&n.isBatchCall)throw new V({code:"BAD_REQUEST",message:"Cannot batch subscription calls"});return[void 0,{data:await o({path:t.path,getRawInput:t.getRawInput,ctx:h.value(),type:o._def.type,signal:e.req.signal})}]}catch(n){var r,a,i;const o=J(n),s=t.result();return null===(r=e.onError)||void 0===r||r.call(e,{error:o,path:t.path,input:s,ctx:h.valueOrUndefined(),type:null!==(a=null===(i=t.procedure)||void 0===i?void 0:i._def.type)&&void 0!==a?a:"unknown",req:e.req}),[o,void 0]}});if(!n.isBatchCall){const[t]=n.calls,[r,a]=await o[0];switch(n.type){case"unknown":case"mutation":case"query":{if(c.set("content-type","application/json"),_t(null==a?void 0:a.data))throw new V({code:"UNSUPPORTED_MEDIA_TYPE",message:"Cannot use stream-like response in non-streaming request - use httpBatchStreamLink"});const o=r?{error:L({config:u,ctx:h.valueOrUndefined(),error:r,input:t.result(),path:t.path,type:n.type})}:{result:{data:a.data}},i=vt({ctx:h.valueOrUndefined(),info:n,responseMeta:e.responseMeta,errors:r?[r]:[],headers:c,untransformedJSON:[o]});return new Response(JSON.stringify(W(u,o)),{status:i.status,headers:c})}case"subscription":{const o=C(()=>{if(r)return gt(r);if(!b)return gt(new V({code:"METHOD_NOT_SUPPORTED",message:'Missing experimental flag "sseSubscriptions"'}));if(!ee(a.data)&&!S(a.data))return gt(new V({message:`Subscription ${t.path} did not return an observable or a AsyncGenerator`,code:"INTERNAL_SERVER_ERROR"}));return ee(a.data)?oe(a.data,e.req.signal):a.data}),i=mt((0,yt.default)((0,yt.default)({},u.sse),{},{data:o,serialize:e=>u.transformer.output.serialize(e),formatError(n){var o,r,a;const i=J(n.error),s=null==t?void 0:t.result(),l=null==t?void 0:t.path,c=null!==(o=null==t||null===(r=t.procedure)||void 0===r?void 0:r._def.type)&&void 0!==o?o:"unknown";null===(a=e.onError)||void 0===a||a.call(e,{error:i,path:l,input:s,ctx:h.valueOrUndefined(),req:e.req,type:c});return L({config:u,ctx:h.valueOrUndefined(),error:i,input:s,path:l,type:c})}}));for(const[e,t]of Object.entries(ft))c.set(e,t);const s=vt({ctx:h.valueOrUndefined(),info:n,responseMeta:e.responseMeta,errors:[],headers:c,untransformedJSON:null});return new Response(i,{headers:c,status:s.status})}}}if("application/jsonl"===n.accept){c.set("content-type","application/json"),c.set("transfer-encoding","chunked");const t=vt({ctx:h.valueOrUndefined(),info:n,responseMeta:e.responseMeta,errors:[],headers:c,untransformedJSON:null}),r=ot((0,yt.default)((0,yt.default)({},u.jsonl),{},{maxDepth:1/0,data:o.map(async t=>{const[o,r]=await t,a=n.calls[0];var i,s;if(o)return{error:L({config:u,ctx:h.valueOrUndefined(),error:o,input:a.result(),path:a.path,type:null!==(i=null===(s=a.procedure)||void 0===s?void 0:s._def.type)&&void 0!==i?i:"unknown"})};const l=ee(r.data)?oe(r.data,e.req.signal):Promise.resolve(r.data);return{result:Promise.resolve({data:l})}}),serialize:e=>u.transformer.output.serialize(e),onError:t=>{var o,r;null===(o=e.onError)||void 0===o||o.call(e,{error:J(t),path:void 0,input:void 0,ctx:h.valueOrUndefined(),req:e.req,type:null!==(r=null==n?void 0:n.type)&&void 0!==r?r:"unknown"})},formatError(e){var t,o;const r=null==n?void 0:n.calls[e.path[0]],a=J(e.error),i=null==r?void 0:r.result(),s=null==r?void 0:r.path,l=null!==(t=null==r||null===(o=r.procedure)||void 0===o?void 0:o._def.type)&&void 0!==t?t:"unknown";return L({config:u,ctx:h.valueOrUndefined(),error:a,input:i,path:s,type:l})}}));return new Response(r,{headers:c,status:t.status})}c.set("content-type","application/json");const r=(await Promise.all(o)).map(e=>{const[t,n]=e;return t?e:_t(n.data)?[new V({code:"UNSUPPORTED_MEDIA_TYPE",message:"Cannot use stream-like response in non-streaming request - use httpBatchStreamLink"}),void 0]:e}),a=r.map(([e,t],o)=>{const r=n.calls[o];var a,i;return e?{error:L({config:u,ctx:h.valueOrUndefined(),error:e,input:r.result(),path:r.path,type:null!==(a=null===(i=r.procedure)||void 0===i?void 0:i._def.type)&&void 0!==a?a:"unknown"})}:{result:{data:t.data}}}),i=r.map(([e])=>e).filter(Boolean),s=vt({ctx:h.valueOrUndefined(),info:n,responseMeta:e.responseMeta,untransformedJSON:a,errors:i,headers:c});return new Response(JSON.stringify(W(u,a)),{status:s.status,headers:c})}catch(t){var w;const[n,o]=f,r=h.valueOrUndefined(),{error:a,untransformedJSON:i,body:s}=function(e,t){const{router:n,req:o,onError:r}=t.opts,a=J(e);null==r||r({error:a,path:t.path,input:t.input,ctx:t.ctx,type:t.type,req:o});const i={error:L({config:n._def._config,error:a,type:t.type,path:t.path,input:t.input,ctx:t.ctx})},s=W(n._def._config,i);return{error:a,untransformedJSON:i,body:JSON.stringify(s)}}(t,{opts:e,ctx:h.valueOrUndefined(),type:null!==(w=null==o?void 0:o.type)&&void 0!==w?w:"unknown"}),l=vt({ctx:r,info:o,responseMeta:e.responseMeta,untransformedJSON:i,errors:[a],headers:c});return new Response(s,{status:l.status,headers:c})}}function St(e){try{var t,n;const o=e.headers[":scheme"]&&"https"===e.headers[":scheme"]||e.socket&&"encrypted"in e.socket&&e.socket.encrypted?"https:":"http:",r=null!==(t=null!==(n=e.headers.host)&&void 0!==n?n:e.headers[":authority"])&&void 0!==t?t:"localhost";return new URL(e.url,`${o}//${r}`)}catch(e){throw new V({code:"BAD_REQUEST",message:"Invalid URL",cause:e})}}function Ct(e){const t=new Headers;for(const n in e){const o=e[n];if("string"!=typeof n||!n.startsWith(":"))if(Array.isArray(o))for(const e of o)t.append(n,e);else null!=o&&t.append(n,o)}return t}function kt(e,t,n){const o=new AbortController,r=()=>{t.off("close",r),e.off("aborted",r),o.abort()};t.once("close",r),e.once("aborted",r);const a=St(e),i={headers:Ct(e.headers),method:e.method,signal:o.signal};"GET"!==e.method&&"HEAD"!==e.method&&(i.body=function(e,t){if("body"in e){if(void 0===e.body)return;return"string"==typeof e.body?e.body:JSON.stringify(e.body)}let n=0,o=!1;return new ReadableStream({start(r){const a=s=>{n+=s.length,!t.maxBodySize||n<=t.maxBodySize?r.enqueue(new Uint8Array(s.buffer,s.byteOffset,s.byteLength)):(r.error(new V({code:"PAYLOAD_TOO_LARGE"})),o=!0,e.off("data",a),e.off("end",i))},i=()=>{o||(o=!0,e.off("data",a),e.off("end",i),r.close())};e.on("data",a),e.on("end",i)},cancel(){e.destroy()}})}(e,n),i.duplex="half");return new Request(a,i)}async function Tt(e){const{res:t}=e;try{const n=new WritableStream({async write(e){var n;await async function(e,t){!1===e.write(t)&&await new Promise((t,n)=>{const o=e=>{n(e),a()},r=()=>{t(),a()},a=()=>{e.off("error",o),e.off("drain",r)};e.once("error",o),e.once("drain",r)})}(t,e),null===(n=t.flush)||void 0===n||n.call(t)}});await e.body.pipeTo(n,{signal:e.signal})}catch(e){if(ue(e))return;throw e}}var Mt=N(U(),1);function Et(e){return t=>{var n;const{res:o,req:r}=e,a=J(t),i=L({config:e.router._def._config,error:a,type:"unknown",path:void 0,input:void 0,ctx:void 0});null===(n=e.onError)||void 0===n||n.call(e,{req:r,error:a,type:"unknown",path:void 0,input:void 0,ctx:void 0});const s=W(e.router._def._config,{error:i});o.statusCode=i.data.httpStatus,o.end(JSON.stringify(s))}}async function $t(e){return new Promise(t=>{var n;const o=null!==(n=e.middleware)&&void 0!==n?n:(e,t,n)=>n();return e.res.once("finish",()=>{t()}),o(e.req,e.res,t=>{C(async()=>{var n;const o=kt(e.req,e.res,{maxBodySize:null!==(n=e.maxBodySize)&&void 0!==n?n:null}),r=await xt((0,Mt.default)((0,Mt.default)({},e),{},{req:o,error:t?J(t):null,createContext:async t=>{var n;return await(null===(n=e.createContext)||void 0===n?void 0:n.call(e,(0,Mt.default)((0,Mt.default)({},e),t)))},onError(t){var n;null==e||null===(n=e.onError)||void 0===n||n.call(e,(0,Mt.default)((0,Mt.default)({},t),{},{req:e.req}))}}));await async function(e){const{response:t,rawResponse:n}=e;200===n.statusCode&&(n.statusCode=t.status);for(const[e,o]of t.headers)n.setHeader(e,o);try{t.body&&await Tt({res:n,signal:e.request.signal,body:t.body})}catch(e){throw n.headersSent||(n.statusCode=500),e}finally{n.end()}}({request:o,response:r,rawResponse:e.res})}).catch(Et(e))})})}const Bt=t("http");var It=N(U(),1);function Ot(e){var t;const n=(null!==(t=e.basePath)&&void 0!==t?t:"/").length;return(t,o)=>{let r="";C(async()=>{const a=St(t);r=a.pathname.slice(n),await $t((0,It.default)((0,It.default)({},e),{},{req:t,res:o,path:r}))}).catch(Et((0,It.default)({req:t,res:o,path:r},e)))}}function Nt(e){return Bt.createServer(function(e){return Ot(e)}(e))}const Pt=t("node:os"),qt=t("node:crypto"),zt=1024**3,Dt=(e,t,n)=>Math.min(Math.max(e,t),n),At=e=>e?40:0,Ft=(e=0)=>{if(!e)return 0;return Dt(e/(12*zt)*20,0,20)},Rt=(e=0)=>{if(!e)return 0;return Dt(e/(32*zt)*10,0,10)},Ut=e=>e?10:0,jt=(e="default",t=null)=>{const n=String(e).toLowerCase();if(!n)return 0;if(n.includes("cuda"))return 20;if(n.includes("vulkan"))return 10;if(n.includes("default")){return"darwin"===t||"ios"===t?15:5}return 0},Lt=({platform:e,variant:t,hasGpu:n,gpuUsableBytes:o=0,cpuUsableBytes:r=0,ok:a=!0}={})=>{if(!a)return 0;const i=At(n)+jt(t,e)+Ft(o),s=Rt(r),l=Ut(a);return Math.min(100,Math.round(i+s+l))},Gt=["cuda","vulkan","snapdragon","default"],Ht=e=>e||0===e?Array.isArray(e)?e.filter(e=>null!=e):[e]:[],Jt=e=>e&&String(e).trim().toLowerCase()||null,Vt=(e={})=>{const t=String(e.type||e.deviceType||e.kind||"").toLowerCase();return!!t.includes("gpu")||(!!t.includes("cuda")||(!!t.includes("metal")||(!!t.includes("vulkan")||!!t.includes("snapdragon"))))},Kt=({platform:e,totalMemoryInBytes:t,variant:n,devices:o,gpuMemoryFraction:r,cpuMemoryFraction:a,ok:i,error:s})=>{const l=(e=>Array.isArray(e)?e.map(e=>({...e})):[])(((e,t)=>"snapdragon"===e?t.filter(e=>"GPUOpenCL"!==e.deviceName):t)(n,o)),c=l.some(Vt),u=l.filter(e=>Vt(e)&&Number.isFinite(Number(e.maxMemorySize))).reduce((e,t)=>e+t.maxMemorySize,0),d=t,p=c?Math.floor(u*r):0,m=d?Math.floor(d*a):0,f={platform:e,variant:n,hasGpu:c,gpuUsableBytes:p,cpuUsableBytes:m,ok:i},h=Lt(f),y=i?(({platform:e,variant:t,hasGpu:n,gpuUsableBytes:o=0,cpuUsableBytes:r=0,ok:a=!0}={})=>({gpuPresence:At(n),variant:jt(t,e),gpuMemory:Ft(o),cpuMemory:Rt(r),availability:Ut(a)}))(f):null;return{platform:e,ok:i,variant:n,hasGpu:c,devices:l,gpuTotalBytes:u,gpuUsableBytes:p,cpuTotalBytes:d,cpuUsableBytes:m,score:h,breakdown:y,error:s,timestamp:(new Date).toISOString()}},Yt=({device:e,modelBytes:t=0,kvCacheBytes:n=0}={})=>{if(!e)return{totalRequiredBytes:t+n,fitsInGpu:!1,fitsInCpu:!1,limiting:"unknown-device"};const o=Math.max(0,Number(t)||0)+Math.max(0,Number(n)||0),r=e.hasGpu&&o>0&&o<=e.gpuUsableBytes,a=o>0&&o<=e.cpuUsableBytes;let i="ok";return!r&&e.hasGpu&&(i="gpu-memory"),a||(i=r?"cpu-memory":"insufficient-memory"),{totalRequiredBytes:o,fitsInGpu:r,fitsInCpu:a,limiting:i}},Wt=async({platform:e,variant:t=null,preferVariants:n=[],variantPreference:o=[],gpuMemoryFraction:r=.85,cpuMemoryFraction:a=.5,includeBreakdown:i=!1,totalMemoryInBytes:s,modelBytes:l=null,kvCacheBytes:c=null,limitedKvCacheBytes:u=null,dependencies:d={},defaultVariants:p=Gt}={})=>{const{getBackendDevicesInfo:m,isLibVariantAvailable:f}=d;if("function"!=typeof m||"function"!=typeof f)throw new TypeError("GGML capability detection requires getBackendDevicesInfo and isLibVariantAvailable functions");const h=(({variant:e,preferVariants:t=[],variantPreference:n=[],defaultVariants:o=Gt}={})=>{const r=[];e&&r.push(e),r.push(...Ht(t)),r.push(...Ht(n)),r.push(...o);const a=r.map(Jt).filter(Boolean);return Array.from(new Set(a))})({variant:t,preferVariants:n,variantPreference:o,defaultVariants:p}),y=[];for(const t of h)try{if(!await f(t))throw new Error(`Variant ${t} not available on this platform`);const n=await m(t);y.push(Kt({platform:e,totalMemoryInBytes:s,variant:t,devices:n,gpuMemoryFraction:r,cpuMemoryFraction:a,ok:!0}))}catch(n){const o=n instanceof Error?n.message:String(n);y.push(Kt({platform:e,totalMemoryInBytes:s,variant:t,devices:[],gpuMemoryFraction:r,cpuMemoryFraction:a,ok:!1,error:o}))}const g=y.filter(e=>e.ok)[0]||null,b={ok:Boolean(g),selected:g?{...g,breakdown:i?g.breakdown:void 0}:null,attempts:y};if(!i&&b.selected&&delete b.selected.breakdown,!b||!l&&!c)return b;const w=e=>{if(!e)return e;const t=Yt({device:e,modelBytes:l||0,kvCacheBytes:c||0});let n=null;return null!=u&&u!==c&&(n=Yt({device:e,modelBytes:l||0,kvCacheBytes:u})),{...e,fit:t,...n&&{limitedFit:n}}};return b.selected=w(b.selected),b.attempts=Array.isArray(b.attempts)?b.attempts.map(w):b.attempts,b},Qt="ggml-llm",Zt=["cuda","vulkan","default"],Xt=new Map([[Qt,Wt],["ggml-stt",async({platform:e,variant:t=null,preferVariants:n=[],variantPreference:o=[],gpuMemoryFraction:r=.85,cpuMemoryFraction:a=.5,includeBreakdown:i=!1,totalMemoryInBytes:s,modelBytes:l=null,processingBytes:c=null,kvCacheBytes:u=null,dependencies:d={}}={})=>{const p=o&&o.length>0?o:Zt;return Wt({platform:e,variant:t,preferVariants:n,variantPreference:p,gpuMemoryFraction:r,cpuMemoryFraction:a,includeBreakdown:i,totalMemoryInBytes:s,modelBytes:l,kvCacheBytes:c??u,dependencies:d,defaultVariants:Zt})}]]),en=async({platform:e,totalMemoryInBytes:t,backend:n=Qt,dependencies:o,...r}={})=>{const a=Xt.get(n);if(!a)throw new Error(`No capability detector registered for backend "${n}"`);return await a({...r,dependencies:o,totalMemoryInBytes:t,platform:e})},tn={f16:2,f32:4,q8_0:1,q6_k:.75,q5_k:.625,q5_k_m:.625,q5_k_s:.625,q5_1:.625,q5_0:.625,q4_k:.5,q4_k_m:.5,q4_k_s:.5,q4_1:.5,q4_0:.5,iq4_nl:.5},nn=e=>{const t=e?String(e).toLowerCase():"f16";return tn[t]||tn.f16},on=({modelBytes:e=0,audioLengthSeconds:t=30,sampleRate:n=16e3,bytesPerSample:o=4}={})=>{const r=Math.max(0,Number(e)||0),a=Math.max(0,Math.floor(Math.max(0,t)*n*o)),i=1048576,s=1024*i;let l,c,u;l=r<200*i?120*i:r<500*i?140*i:r<2*s?150*i:160*i,c=r<200*i?70*i:r<500*i?135*i:220*i,u=r<100*i?20*i:r<200*i?30*i:r<500*i?85*i:r<2*s?215*i:360*i;const d=l+c+u;return{modelBytes:r,audioBufferBytes:a,processingBufferBytes:d,totalBytes:r+d+a}},rn=e=>e?String(e).trim().toLowerCase():null,an=({arch:e,nLayer:t=0})=>({arch:rn(e),enabled:!1,window:null,pattern:null,denseFirst:!1,type:null,kvLayers:Math.max(0,Math.floor(Number(t)||0)),swaLayers:0}),sn=new Map([["llama4",({nSwa:e})=>{if(0===e)return{enabled:!1};return{enabled:!0,window:e&&e>0?e:8192,pattern:4,type:"chunked"}}],["afmoe",({nSwa:e})=>!e||e<=0?{enabled:!1}:{enabled:!0,window:e,pattern:4,type:"standard"}],["phi3",()=>({enabled:!1})],["gemma2",({nSwa:e})=>{const t=e&&e>0?e:4096;return t?{enabled:!0,window:t,pattern:2,type:"standard"}:{enabled:!1}}],["gemma3",({nSwa:e})=>!e||e<=0?{enabled:!1}:{enabled:!0,window:e,pattern:6,type:"standard"}],["gemma3n",({nLayer:e,nSwa:t})=>!t||t<=0?{enabled:!1}:{enabled:!0,window:t,pattern:5,type:"standard",kvLayers:Math.min(20,e)}],["gemma-embedding",({nSwa:e})=>!e||e<=0?{enabled:!1}:{enabled:!0,window:e,pattern:6,type:"symmetric"}],["cohere2",({nSwa:e})=>!e||e<=0?{enabled:!1}:{enabled:!0,window:e,pattern:4,type:"standard"}],["olmo2",({nSwa:e})=>!e||e<=0?{enabled:!1}:{enabled:!0,window:e,pattern:4,type:"standard"}],["exaone4",({nLayer:e,nSwa:t})=>{let n=null;return t&&t>0?n=t:e>=64&&(n=4096),n?{enabled:!0,window:n,pattern:4,type:"standard"}:{enabled:!1}}],["gpt-oss",({nSwa:e})=>!e||e<=0?{enabled:!1}:{enabled:!0,window:e,pattern:2,type:"standard"}],["smallthinker",({nSwa:e})=>!e||e<=0?{enabled:!1}:{enabled:!0,window:4096,pattern:4,denseFirst:!0,type:"standard"}]]),ln=({arch:e,metadata:t={},nLayer:n=0}={})=>{const o=rn(e||t["general.architecture"]),r=Math.max(0,Math.floor(Number(n)||0)),a=((e={},t=null)=>{if(!e)return null;const n=rn(t),o=n?`${n}.attention.sliding_window`:null,r=(o&&null!=e[o]?e[o]:null)??e["llama.attention.sliding_window"];if(null==r)return null;const a=Number(r);return Number.isFinite(a)?a:null})(t,o),i=o?sn.get(o):null;if(!i)return an({arch:o,nLayer:n});const s=i({nLayer:r,nSwa:a,metadata:t});if(!s||!s.enabled||!s.window||s.window<=0)return an({arch:o,nLayer:n});const l=Math.max(0,Math.floor(Number(s.pattern)||0)),c=null!=s.kvLayers&&Number.isFinite(Number(s.kvLayers))?Number(s.kvLayers):r,u=Math.max(0,Math.floor(c)),d=((e=0,t=0,n=!1)=>{const o=Math.max(0,Math.floor(Number(e)||0)),r=Math.max(0,Math.floor(Number(t)||0));if(!o||1===r)return 0;if(r<=0)return o;const a=Math.max(0,r-1),i=o%r;return Math.floor(o/r)*a+(n?Math.max(0,i-1):Math.min(i,a))})(u,l,Boolean(s.denseFirst));return{arch:o,enabled:d>0,window:s.window,pattern:l,denseFirst:Boolean(s.denseFirst),type:s.type||"standard",kvLayers:u,swaLayers:d}},cn=(e={})=>{const t=e["general.architecture"],n=(t,n=null)=>{const o=e[t],r=Number(o);return Number.isFinite(r)?r:n},o=t?n(`${t}.context_length`,n("llama.context_length")):null,r=t?n(`${t}.block_count`,n("llama.block_count")):null,a=t?n(`${t}.embedding_length`,n("llama.embedding_length")):null,i=t?n(`${t}.attention.head_count`,n("llama.attention.head_count")):null;return{arch:t,nCtxTrain:o,nLayer:r,nEmbd:a,nHead:i,nHeadKv:t?n(`${t}.attention.head_count_kv`,n("llama.attention.head_count_kv",i)):null,nEmbdHeadK:t?n(`${t}.attention.key_length`,n("llama.attention.key_length")):null,nEmbdHeadV:t?n(`${t}.attention.value_length`,n("llama.attention.value_length")):null,quantVersion:e["general.quantization_version"]||null,fileType:e["general.file_type"]||null}},un=({layerCount:e,headKvCount:t,embdHeadKCount:n,embdHeadVCount:o,cacheTypes:r,swaConfig:a,kvUnified:i=!1,nParallel:s=1,swaFull:l=!1})=>{const c=a?.window&&i?Math.max(1,Number(s)||1):1;return i=>((e,t,n,o,r,a={},{totalLayers:i=null,swaLayers:s=0,swaContext:l=null,swaContextMultiplier:c=1,swaAdditionalTokens:u=0,swaFull:d=!1}={})=>{if(!(e&&t&&n&&o&&r))return 0;const p=Number(null!=i&&void 0!==i?i:e),m=Math.max(0,Math.floor(p));if(!m)return 0;const f=nn(a.k),h=nn(a.v),y=Number(n)*(Number(o)*f+Number(r)*h);if(!y)return 0;const g=Math.max(0,Number(t)||0),b=Math.min(m,Math.max(0,Math.floor(Number(s)||0))),w=Math.max(0,m-b),v=(null!=l&&Number.isFinite(Number(l))?Math.max(0,Number(l)):g)*Math.max(1,Number(c)||1)+Math.max(0,Number(u)||0),_=d?g:Math.min(g,v),x=w*g+b*Math.max(0,Math.floor(_));return Math.round(y*x)})(e,i,t,n,o,r,{totalLayers:e,swaLayers:a?.swaLayers||0,swaContext:a?.window,swaFull:l,swaContextMultiplier:c})},dn=({maxCtx:e,availableMemory:t,modelBytes:n,kvBytesForCtx:o})=>{const r=Math.max(1,Math.floor(Number(e)||0));if(!o||t<=n)return r;let a=1,i=r,s=r;for(;a<=i;){const e=Math.floor((a+i)/2);n+o(e)<=t?(s=e,a=e+1):i=e-1}return s};var pn=f(285);const mn=t("node:stream/web");var fn=f.t(mn,2);const hn=t("node:events"),yn=new hn.EventEmitter;yn.setMaxListeners(100);const gn=(e,t,n)=>{e.push({...t,timestamp:t.timestamp||(new Date).toISOString()}),e.length>n&&e.shift()};class bn{constructor(e=9999){this.maxEntries=e,this.modelLoads=[],this.completions=[],this.transcriptions=[]}addModelLoad(e){gn(this.modelLoads,e,this.maxEntries),yn.emit("status:modelLoad",e),yn.emit("status:change",{type:"modelLoad",entry:e})}addCompletion(e){gn(this.completions,e,this.maxEntries),yn.emit("status:completion",e),yn.emit("status:change",{type:"completion",entry:e})}addTranscription(e){gn(this.transcriptions,e,this.maxEntries),yn.emit("status:transcription",e),yn.emit("status:change",{type:"transcription",entry:e})}getModelLoadHistory(){return[...this.modelLoads].reverse()}getCompletionHistory(){return[...this.completions].reverse()}getTranscriptionHistory(){return[...this.transcriptions].reverse()}clear(){this.modelLoads=[],this.completions=[],this.transcriptions=[]}}const wn=new bn,vn=new bn;let _n=0;function xn(e){const t=t=>e(t);return yn.on("status:change",t),()=>yn.off("status:change",t)}function Sn(e){const t=[];return{generators:Array.from(e.entries()).filter(([,e])=>"ggml-llm"===e.type).map(([e,n])=>{const{instance:o}=n;let r=[];return o.contexts&&(r=Array.from(o.contexts.entries()).map(([n,o])=>{const r={key:n,refCount:o.refCount,hasModel:Boolean(o.context)},a=o.context.parallel.getStatus();return r.parallelStatus=a,t.push({generatorId:e,contextKey:n,...a}),r})),{id:e,type:n.type,refCount:n.refCount,repoId:o.info?.model?.repoId||null,quantization:o.info?.model?.quantization||null,variant:o.info?.runtime?.variant||null,nCtx:o.info?.runtime?.n_ctx||null,nParallel:o.info?.runtime?.n_parallel||null,contexts:r}}),parallelStatuses:t,history:{modelLoads:wn.getModelLoadHistory(),completions:wn.getCompletionHistory()}}}function Cn(e){return{generators:Array.from(e.entries()).filter(([,e])=>"ggml-stt"===e.type).map(([e,t])=>{const{instance:n}=t,o=n.getStatus?.()||{},r=o.queueStatus||{processing:!1,queuedCount:0};return{id:e,type:t.type,refCount:t.refCount,repoId:n.info?.model?.repoId||null,quantization:n.info?.model?.quantization||null,variant:n.info?.runtime?.variant||null,hasContext:o.hasContext||!1,contextRefCount:o.contextRefCount||0,queueStatus:r}}),history:{modelLoads:vn.getModelLoadHistory(),transcriptions:vn.getTranscriptionHistory()}}}const kn="undefined"!=typeof globalThis&&globalThis.ReadableStream&&globalThis.WritableStream?{ReadableStream:globalThis.ReadableStream,WritableStream:globalThis.WritableStream}:fn,{ReadableStream:Tn,WritableStream:Mn}=kn,En=new TextEncoder,$n=(e={},t={})=>(Object.entries(t||{}).forEach(([t,n])=>{n&&"object"==typeof n&&!Array.isArray(n)?(e[t]&&"object"==typeof e[t]||(e[t]={}),$n(e[t],n)):e[t]=n}),e),Bn="https://huggingface.co",In="https://huggingface.co/api",On=g.join(Pt.homedir(),".buttress","models"),Nn=["mxfp4","q8_0","q6_k","q6","q5_k_m","q5_k_s","q5_k","q5_1","q5_0","q4_k_m","q4_k_s","q4_k","q4_1","q4_0","q3","q2"],Pn=.5,qn={backend:{type:"ggml-llm",variant:null,variant_preference:["cuda","vulkan","snapdragon","default"],gpu_memory_fraction:.85,cpu_memory_fraction:Pn},model:{repo_id:null,revision:"main",filename:null,url:null,quantization:null,preferred_quantizations:[],n_ctx:null,n_gpu_layers:"auto",n_parallel:4,n_batch:2048,flash_attn:"auto",cache_type_k:"f16",cache_type_v:"f16",ctx_shift:!1,kv_unified:!1,swa_full:!1,allow_local_file:!1,local_path:null,api_base:In,base_url:Bn},runtime:{cache_dir:On,prefer_variants:[],huggingface_token:process.env.HUGGINGFACE_TOKEN||null,http_headers:{},session_cache:{enabled:!0,max_size_bytes:10737418240,max_entries:1e3},context_release_delay_ms:1e4}},zn=(e,t=[])=>e||0===e?Array.isArray(e)?e.filter(e=>null!=e):[e]:[...t],Dn=e=>{if(!e)return null;const t=String(e).toLowerCase();return["cuda","vulkan","snapdragon","default"].includes(t)?t:null},An=(e={})=>{const t=JSON.parse(JSON.stringify(qn));if($n(t,e),t.backend.variant=Dn(t.backend.variant),t.backend.variant_preference=Array.from(new Set(zn(t.backend.variant_preference).map(Dn).filter(Boolean))),0===t.backend.variant_preference.length&&(t.backend.variant_preference=["cuda","vulkan","snapdragon","default"]),t.runtime.prefer_variants=Array.from(new Set(zn(t.runtime.prefer_variants).map(Dn).filter(Boolean))),t.model.preferred_quantizations=Array.from(new Set(zn(t.model.preferred_quantizations||t.model.quantizations).map(e=>e?String(e).toLowerCase():null).filter(Boolean))),t.model.quantization){const e=String(t.model.quantization).toLowerCase();t.model.preferred_quantizations.includes(e)||t.model.preferred_quantizations.unshift(e)}return t.model.n_parallel=Math.max(1,Number(t.model.n_parallel)||4),t.model.n_batch=Math.max(1,Number(t.model.n_batch)||512),t.model.base_url=t.model.base_url||Bn,t.model.api_base=t.model.api_base||In,t.runtime.cache_dir=t.runtime.cache_dir?g.resolve(t.runtime.cache_dir):On,t.runtime.session_cache={...qn.runtime.session_cache,...t.runtime.session_cache||{}},t.runtime.context_release_delay_ms=Math.max(0,Number(t.runtime.context_release_delay_ms)||qn.runtime.context_release_delay_ms),t},Fn=e=>{const t=e.toLowerCase();return Nn.find(e=>t.includes(e))||null},Rn=async e=>{await(0,y.mkdir)(e,{recursive:!0})},Un=(e,t,n=On)=>{const o=(0,qt.createHash)("sha256").update(e).digest("hex");return g.join(((e=On)=>g.join(e,".metadata-cache"))(n),t,`${o}.json`)},jn=async(e,t,n=On)=>{try{const o=Un(e,t,n),r=await(0,y.readFile)(o,"utf-8");return console.log(`[Cache] Hit ${t} cache:`,g.basename(o)),JSON.parse(r,(e,t)=>"string"==typeof t&&t.startsWith("__bigint__")?BigInt(t.slice(10)):t)}catch(e){return null}},Ln=async(e,t,n,o=On)=>{try{const r=Un(e,t,o);await Rn(g.dirname(r)),await(0,y.writeFile)(r,JSON.stringify(n,(e,t)=>"bigint"==typeof t?`__bigint__${t.toString()}`:t),"utf-8"),console.log(`[Cache] Wrote ${t} cache:`,g.basename(r))}catch(e){console.warn(`[Cache] Failed to write ${t} cache:`,e.message)}},Gn=(e=On)=>g.join(e,".session-state-cache"),Hn=(e=On)=>g.join(Gn(e),"cache-map.json"),Jn=(e=On)=>g.join(Gn(e),"temp"),Vn=(e=On)=>g.join(Gn(e),"states"),Kn=async(e,t=On)=>{const n=Hn(t),o=`${n}.tmp.${Date.now()}`;try{await Rn(g.dirname(n)),await(0,y.writeFile)(o,JSON.stringify(e,null,2),"utf-8"),await(0,y.rename)(o,n)}catch(e){throw await(0,y.unlink)(o).catch(()=>{}),e}},Yn=(e,t,n)=>{const o=Object.values(n.entries);console.log(`[SessionCache] Finding match for promptText (${e.length} chars)`),console.log(`[SessionCache] Checking ${o.length} cache entries`);const r=o.filter(e=>{return n=e.metadata,o=t,n.modelPath===o.modelPath&&n.variant===o.variant&&n.n_gpu_layers===o.n_gpu_layers&&n.n_ctx>=o.n_ctx&&n.cacheTypeK===o.cacheTypeK&&n.cacheTypeV===o.cacheTypeV;var n,o}).reduce((t,n)=>{const o=((e,t)=>{const n=Math.min(e.length,t.length);let o=0;for(;o<n&&e[o]===t[o];)o+=1;return o})(e,n.fullText);return o>=100&&o>t.prefixLen?{entry:n,prefixLen:o}:t},{entry:null,prefixLen:0});return r.entry?(console.log(`[SessionCache] Prefix match found: ${r.entry.id} (${r.prefixLen}/${r.entry.fullText.length} chars)`),{entry:r.entry,prefixLength:r.prefixLen}):(console.log("[SessionCache] No match found"),null)};class Wn{constructor(e,t){this.config=e,this.plan=t,this.baseDir=e.runtime.cache_dir,this.enabled=!1!==e.runtime.session_cache?.enabled,this.maxSizeBytes=((e,t)=>{if(null==e)return t;if("number"==typeof e)return e;if("string"==typeof e){const n=pn.parse(e);return null!=n?n:t}return t})(e.runtime.session_cache?.max_size_bytes,10737418240),this.maxEntries=e.runtime.session_cache?.max_entries||1e3,this.metadata={variant:t.info?.runtime?.variant||null,n_gpu_layers:t.info?.runtime?.n_gpu_layers||0,n_ctx:t.info?.runtime?.n_ctx||0,modelPath:t.localPath,cacheTypeK:t.info?.runtime?.cache_type_k||"f16",cacheTypeV:t.info?.runtime?.cache_type_v||"f16"},this.cacheMap=null,this.initialized=!1}async initialize(){if(this.enabled&&!this.initialized)try{await Rn(Gn(this.baseDir)),await Rn(Jn(this.baseDir)),await Rn(Vn(this.baseDir)),this.cacheMap=await(async(e=On)=>{try{const t=Hn(e),n=await(0,y.readFile)(t,"utf-8"),o=JSON.parse(n);return o.entries&&"object"==typeof o.entries?o:{version:1,entries:{},totalSize:0}}catch{return{version:1,entries:{},totalSize:0}}})(this.baseDir),this.initialized=!0,console.log(`[SessionCache] Initialized with ${Object.keys(this.cacheMap.entries).length} entries`)}catch(e){console.warn(`[SessionCache] Failed to initialize: ${e.message}`),this.enabled=!1}}async findMatchingEntry(e){if(!this.enabled||!this.cacheMap)return null;const t=Yn(e,this.metadata,this.cacheMap);if(t){const{entry:e}=t;return await(async e=>{try{return await(0,y.stat)(e),!0}catch{return!1}})(e.stateFilePath)?(e.lastAccessedAt=(new Date).toISOString(),await Kn(this.cacheMap,this.baseDir).catch(()=>{}),{entry:e}):(console.log(`[SessionCache] Removing stale entry: ${e.id}`),delete this.cacheMap.entries[e.id],this.cacheMap.totalSize-=e.stateFileSize||0,await Kn(this.cacheMap,this.baseDir).catch(()=>{}),null)}return null}async prepareCompletionOptions(e,t){if(!this.enabled)return{options:e,cacheEntry:null,promptPrefix:null};const n=await this.findMatchingEntry(t);if(n){const{entry:t}=n;console.log(`[SessionCache] Found matching entry: ${t.id} (${t.fullText.length} chars, loadStateSize=${t.loadStateSize})`);return{options:{...e,load_state_path:t.stateFilePath},cacheEntry:t,promptPrefix:t.fullText}}return{options:e,cacheEntry:null,promptPrefix:null}}async saveCompletionState(e,t,n,o=0){if(!this.enabled)return null;const r=((e,t)=>{const n=JSON.stringify({text:e,model:t.modelPath,variant:t.variant,n_gpu_layers:t.n_gpu_layers,n_ctx:t.n_ctx});return(0,qt.createHash)("sha256").update(n).digest("hex").slice(0,24)})(e,this.metadata);if(this.cacheMap.entries[r])return console.log(`[SessionCache] Entry already exists for prompt: ${r}`),await(0,y.unlink)(n).catch(()=>{}),this.cacheMap.entries[r];const a=e+t,i=((e,t=On)=>g.join(Vn(t),`${e}.bin`))(r,this.baseDir);try{await Rn(g.dirname(i)),await(0,y.rename)(n,i);const s=await(0,y.stat)(i),l={id:r,promptText:e,completionText:t,fullText:a,promptTokenCount:o,stateFilePath:i,stateFileSize:s.size,metadata:{...this.metadata},createdAt:(new Date).toISOString(),lastAccessedAt:(new Date).toISOString()};return this.cacheMap.entries[r]=l,this.cacheMap.totalSize+=s.size,await(async(e,t,n)=>{const o=Object.values(e.entries).sort((e,t)=>new Date(e.lastAccessedAt)-new Date(t.lastAccessedAt));let r=e.totalSize,a=Object.keys(e.entries).length;const i=o.filter(e=>(r>t||a>n)&&(r-=e.stateFileSize||0,a-=1,!0));return await Promise.all(i.map(async t=>{await(0,y.unlink)(t.stateFilePath).catch(()=>{}),delete e.entries[t.id],console.log(`[SessionCache] Evicted entry: ${t.id}`)})),e.totalSize=Math.max(0,r),i.map(e=>e.id)})(this.cacheMap,this.maxSizeBytes,this.maxEntries),await Kn(this.cacheMap,this.baseDir),console.log(`[SessionCache] Saved entry: ${r} (${s.size} bytes)`),l}catch(e){return console.warn(`[SessionCache] Failed to save state: ${e.message}`),await(0,y.unlink)(n).catch(()=>{}),null}}async generateTempStatePath(){return await Rn(Jn(this.baseDir)),((e=On)=>{const t=`${Date.now()}-${Math.random().toString(36).slice(2,10)}`;return g.join(Jn(e),`${t}.bin`)})(this.baseDir)}async cleanup(){await(async(e=On)=>{const t=Jn(e);try{const e=await(0,y.readdir)(t),n=Date.now(),o=36e5;await Promise.all(e.map(async e=>{const r=g.join(t,e),a=await(0,y.stat)(r).catch(()=>null);a&&n-a.mtimeMs>o&&(await(0,y.unlink)(r).catch(()=>{}),console.log(`[SessionCache] Cleaned up temp file: ${e}`))}))}catch{}})(this.baseDir)}}const Qn=async(e,t={})=>{if("function"!=typeof fetch)throw new Error("Global fetch is not available in this runtime");const n=await fetch(e,t);if(!n.ok){const t=await n.text().catch(()=>"");throw new Error(`Failed to fetch ${e}: ${n.status} ${n.statusText} ${t}`.trim())}return n.json()},Zn=async(e,t={})=>{if("function"!=typeof fetch)throw new Error("Global fetch is not available in this runtime");const n=await fetch(e,{...t,method:"HEAD"});if(!n.ok)throw new Error(`Failed to fetch headers for ${e}: ${n.status} ${n.statusText}`);return n},Xn=async(e,t,n=On)=>{const o=JSON.stringify({url:e,headers:t}),a=await jn(o,"range-metadata",n);if(a)return a;const i=!/^https?:/i.test(e),{metadata:s}=await r(e,{fetch,additionalFetchHeaders:t,allowLocalFile:i});return await Ln(o,"range-metadata",s,n),s},eo=(e,t)=>{if(e.model.local_path)return g.resolve(e.model.local_path);const n=t.repoId.split("/"),o=g.join(e.runtime.cache_dir,...n,t.revision);return g.join(o,t.filename)},to=async(e,t)=>{try{const n=await(0,y.stat)(e);return!t||n.size===t}catch(e){return!1}},no=async(e,t,n,o,r)=>{if("function"!=typeof fetch)throw new Error("Global fetch is not available in this runtime");await Rn(g.dirname(n));const a=await fetch(e,{headers:t});if(!a.ok||!a.body)throw new Error(`Failed to download ${e}: ${a.status} ${a.statusText}`);const i=await(0,y.open)(n,"w"),s=Number(a.headers.get("content-length"))||o||0;let l=0,c=.05;try{await a.body.pipeTo(new Mn({async write(e){if(await i.write(e),l+=e.byteLength,"function"==typeof r&&s>0){const e=Math.min(1,l/s);for(;e>=c;)r(c),c+=.05}},async close(){await i.close(),"function"==typeof r&&r(1)},async abort(e){throw await i.close().catch(()=>{}),await(0,y.unlink)(n).catch(()=>{}),e}}))}catch(e){throw await i.close().catch(()=>{}),await(0,y.unlink)(n).catch(()=>{}),e}if(o){const e=await(0,y.stat)(n);if(e.size!==o)throw await(0,y.unlink)(n).catch(()=>{}),new Error(`Downloaded file size mismatch, expected ${o} got ${e.size}`)}},oo=async e=>{const t=e.model.repo_id||e.model.repository||e.model.model;if(!t)throw new Error("`model.repo_id` is required in Buttress backend config");const n=e.model.revision||"main",o=e.runtime.cache_dir,r=JSON.stringify({repoId:t,revision:n,filename:e.model.filename,url:e.model.url,quantization:e.model.quantization,preferred_quantizations:e.model.preferred_quantizations}),a=await jn(r,"artifact-info",o);if(a)return a;const i={...e.runtime.http_headers||{}};if(e.runtime.huggingface_token&&(i.Authorization=`Bearer ${e.runtime.huggingface_token}`),e.model.url){const a=await Zn(e.model.url,{headers:i}),s=Number(a.headers.get("content-length"))||null,l={repoId:t,revision:n,filename:e.model.filename||e.model.url.split("/").pop(),url:e.model.url,size:s,headers:i};return await Ln(r,"artifact-info",l,o),l}let{filename:s}=e.model,l=e.model.quantization&&String(e.model.quantization).toLowerCase();const c=await Qn(`${e.model.api_base}/models/${t}?revision=${n}&blobs=true`,{headers:i}),u=(c?.siblings||c?.files||[]).map(e=>e.rfilename||e.path||e.filename).filter(e=>"string"==typeof e&&e.endsWith(".gguf"));if(0===u.length)throw new Error(`No GGUF artifacts found in repo ${t}`);const d=e.model.preferred_quantizations.length>0?e.model.preferred_quantizations:Nn;if(s)l||(l=Fn(s));else{const e=(()=>{const e=d.find(e=>u.find(t=>t.toLowerCase().includes(e)));if(e)return{filename:u.find(t=>t.toLowerCase().includes(e)),quantization:e};return null})()||{filename:u[0],quantization:null},{filename:t,quantization:n}=e;s=t,l=n||Fn(s)}const p=`${e.model.base_url.replace(/\/+$/,"")}/${t}/resolve/${n}/${s}`,m=/-(\d{5})-of-(\d{5})\.gguf$/,f=s.match(m);let h=null;if(f){const[,,o]=f,r=await Qn(`${e.model.api_base}/models/${t}?revision=${n}&blobs=true`,{headers:i}),a=r?.siblings||r?.files||[],l=Number(o);h=0;for(let e=1;e<=l;e+=1){const t=String(e).padStart(5,"0"),n=s.replace(m,`-${t}-of-${o}.gguf`),r=a.find(e=>(e.rfilename||e.path||e.filename)===n),i=Number(r?.size);Number.isFinite(i)&&i>0&&(h+=i)}}else{const e=await Zn(p,{headers:i});h=Number(e.headers.get("content-length"))||null}const y={repoId:t,revision:n,filename:s,url:p,size:h,quantization:l,headers:i,isSplit:Boolean(f),splitCount:f?Number(f[2]):0};return await Ln(r,"artifact-info",y,o),y},ro=async(e,{modelBytes:t=null,kvCacheBytes:n=null}={})=>{const o=(e=>{const t=[];return e.backend.variant&&t.push(e.backend.variant),e.runtime.prefer_variants.length>0&&t.push(...e.runtime.prefer_variants),t.push(...e.backend.variant_preference),t.push("default"),Array.from(new Set(t.map(Dn).filter(Boolean)))})(e),[r,...s]=o,l=null!=e.backend?.gpu_memory_fraction?Math.min(1,Math.max(0,Number(e.backend.gpu_memory_fraction))):qn.backend.gpu_memory_fraction||1,c=null!=e.backend?.cpu_memory_fraction?Math.min(1,Math.max(0,Number(e.backend.cpu_memory_fraction))):Pn,u=await en({platform:process.platform,totalMemoryInBytes:Pt.totalmem(),backend:"ggml-llm",variant:r||null,preferVariants:s,gpuMemoryFraction:l,cpuMemoryFraction:c,dependencies:{getBackendDevicesInfo:a,isLibVariantAvailable:i},modelBytes:t,kvCacheBytes:n}),d=e=>({...e,devices:Array.isArray(e.devices)?e.devices:[],ok:e.ok,hasGpu:Boolean(e.hasGpu),totalMemory:e.gpuTotalBytes||e.totalMemory||0,error:e.ok?null:new Error(e.error||`Variant ${e.variant} not available on this platform`)});if(!u.ok||!u.selected){const e=(u.attempts||[]).map(e=>`${e.variant}: ${e.error||"unknown error"}`).join("; ");throw new Error(`Unable to initialize any backend variant (${o.join(", ")}). Errors: ${e}`)}const p=(u.attempts||[]).map(d);return{selected:d(u.selected),attempts:p}},ao=async e=>{const t=await oo(e),n=await Xn(t.url,t.headers,e.runtime.cache_dir),{arch:o,nCtxTrain:r,nLayer:a,nEmbd:i,nHead:s,nHeadKv:l,nEmbdHeadK:c,nEmbdHeadV:u,quantVersion:d,fileType:p}=cn(n),m=Number.isFinite(Number(a))?Number(a):0,f=Number.isFinite(Number(i))?Number(i):0,h=Number.isFinite(Number(s))?Number(s):0,y=Number.isFinite(Number(l))?Number(l):h,g=h>0&&f>0?f/h:128,b=null!=c&&Number.isFinite(Number(c))?Number(c):g,w=null!=u&&Number.isFinite(Number(u))?Number(u):g,v=ln({arch:o,metadata:n,nLayer:m}),_=v&&Number.isFinite(Number(v.kvLayers))?Number(v.kvLayers):m,x=Math.max(0,Math.floor(Number(_)||0)),S=e.model.n_ctx?Number(e.model.n_ctx):null;let C=S||r||4096;const k=[],T=[];let M=!0;if(S&&r&&S>r){M=!1;const e=`Requested context length (${S}) exceeds model training context (${r})`;k.push(e),T.push(e),C=r}S&&!r&&k.push("Model metadata missing training context length, using requested value");const E={k:e.model.cache_type_k,v:e.model.cache_type_v},$=t.size>0?t.size:0,B=un({layerCount:x,headKvCount:y,embdHeadKCount:b,embdHeadVCount:w,cacheTypes:E,swaConfig:v,kvUnified:e.model.kv_unified,nParallel:e.model.n_parallel,swaFull:e.model.swa_full}),I=B(C),O=await ro(e,{modelBytes:$,kvCacheBytes:I}),N=O.selected.totalMemory||0,P=N*(e.backend.gpu_memory_fraction||1),q=null!=e.backend.cpu_memory_fraction?Math.min(1,Math.max(0,Number(e.backend.cpu_memory_fraction))):Pn,z=Math.max(0,Pt.totalmem()*q),D=O.selected.hasGpu?P:z,A=dn({maxCtx:C,availableMemory:D,modelBytes:$,kvBytesForCtx:B});if(!S&&A){const e=r?Math.min(A,r):A,t=Math.max(32,e);t<C&&k.push(`Context length capped to ${t} by memory limits`),C=t}C>A&&(C=A);const F=Math.floor(A);console.log(`[buttress] Memory-limited context length: ${F}`);const R=B(C),U=$+R,j=m?$/(m+1):$;let L,G=0;O.selected.hasGpu&&j>0&&(G=Math.min(m+1,Math.max(0,Math.floor(P/j)))),console.log(`[buttress] Auto GPU layer capacity (${O.selected.variant}): ${G}/${m+1}`),L="auto"===e.model.n_gpu_layers||null==e.model.n_gpu_layers?G:Math.max(0,Math.min(Number(e.model.n_gpu_layers)||0,m+1));const H=(()=>{const t=e.model.flash_attn&&String(e.model.flash_attn).toLowerCase();return"on"===t||"off"===t?t:O.selected.hasGpu?"auto":"off"})(),J=e.runtime.cache_dir,V=eo(e,t),K=await to(V,t.size),Y={ok:M,backend:"ggml-llm",warnings:k,errors:T,model:{repoId:t.repoId,revision:t.revision,filename:t.filename,quantization:t.quantization,url:t.url,sizeBytes:t.size,metadata:{architecture:o,n_ctx_train:r,n_layer:m,n_embd:f,quantization_version:d,file_type:p,kv_layer_count:x,swa:v?.enabled?{window:v.window,pattern:v.pattern,dense_first:v.denseFirst,type:v.type,layers:v.swaLayers}:null}},runtime:{variant:O.selected.variant,n_ctx:C,requested_ctx:S,n_gpu_layers:L,n_parallel:e.model.n_parallel,n_batch:e.model.n_batch,flash_attn_type:H,cache_type_k:E.k,cache_type_v:E.v,ctx_shift:e.model.ctx_shift,kv_unified:e.model.kv_unified,swa_full:e.model.swa_full,estimated_max_n_ctx:F,auto_gpu_layers:G},resources:{modelBytes:$,kvCacheBytes:R,totalEstimatedBytes:U,gpuCapacityBytes:N,gpuUsableBytes:P,cpuUsableBytes:z,fit:O.selected.fit},devices:{selected:O.selected,attempts:O.attempts},download:{cacheDir:J,localPath:V,exists:K},timestamp:(new Date).toISOString()};return{config:e,info:Y,artifact:t,metadata:{arch:o,nCtxTrain:r,nLayer:m,nEmbd:f},devices:O,cacheTypes:E,localPath:V,localExists:K}},io=(e,t)=>`event: ${e}\ndata: ${null==t?"":JSON.stringify(t)}\n\n`,so=(e,t,n=null,o=null)=>{let r;const a=Date.now();let i=0;return new Tn({async start(s){try{const l=await e.parallel.completion(t,(e,t)=>{t&&(t.token&&(i+=1),s.enqueue(En.encode(io("token",{requestId:e,...t}))))}),{requestId:c}=l;r=l.stop;const u=await l.promise;console.log("[Completion] Result:",u),s.enqueue(En.encode(io("result",{requestId:c,...u}))),s.close();const d=Date.now()-a,p=u.timings||{};wn.addCompletion({id:`completion-${c}`,generatorId:n,requestId:c,repoId:o?.repoId||null,quantization:o?.quantization||null,variant:o?.variant||null,promptTokens:p.prompt_n??0,tokensGenerated:p.predicted_n??i,tokensPerSecond:p.predicted_per_second??0,promptPerSecond:p.prompt_per_second??0,durationMs:d,success:!0,interrupted:u.interrupted||!1,contextFull:u.context_full||u.contextFull||!1})}catch(e){s.enqueue(En.encode(io("error",{message:e?.message||String(e)}))),s.error(e),wn.addCompletion({id:`completion-${Date.now()}`,generatorId:n,repoId:o?.repoId||null,quantization:o?.quantization||null,variant:o?.variant||null,durationMs:Date.now()-a,tokensGenerated:i,success:!1,error:e?.message||String(e)})}},cancel(){r&&r()}})},lo=(e,t,n,o,r,a,i=null,s=null)=>{let l,c="",u=!1;const d=Date.now();let p=0;return new Tn({async start(m){try{const f=await e.parallel.completion(t,(e,t)=>{t&&(t.token&&(c+=t.token,p+=1),m.enqueue(En.encode(io("token",{requestId:e,...t}))))}),{requestId:h}=f;l=f.stop;const g=await f.promise;g.text?c=g.text:g.content&&(c=g.content),u=!g.interrupted&&!g.context_full,console.log("[Completion] Result:",g),m.enqueue(En.encode(io("result",{requestId:h,...g}))),m.close();const b=Date.now()-d,w=g.timings||{};wn.addCompletion({id:`completion-${h}`,generatorId:i,requestId:h,repoId:s?.repoId||null,quantization:s?.quantization||null,variant:s?.variant||null,promptTokens:w.prompt_n??a??0,tokensGenerated:w.predicted_n??p,tokensPerSecond:w.predicted_per_second??0,promptPerSecond:w.prompt_per_second??0,durationMs:b,success:!0,interrupted:g.interrupted||!1,contextFull:g.context_full||g.contextFull||!1,usedCache:Boolean(t.load_state_path)}),u&&n.enabled&&c?n.saveCompletionState(o,c,r,a).catch(e=>{console.warn("[SessionCache] Save failed:",e.message)}):r&&(0,y.unlink)(r).catch(()=>{})}catch(e){m.enqueue(En.encode(io("error",{message:e?.message||String(e)}))),m.error(e),wn.addCompletion({id:`completion-${Date.now()}`,generatorId:i,repoId:s?.repoId||null,quantization:s?.quantization||null,variant:s?.variant||null,durationMs:Date.now()-d,tokensGenerated:p,success:!1,error:e?.message||String(e)}),(0,y.unlink)(r).catch(()=>{})}},cancel(){l&&l(),(0,y.unlink)(r).catch(()=>{})}})},co=e=>{const t={model:e.plan.localPath,runtime:e.plan.info.runtime};return(0,qt.createHash)("sha256").update(JSON.stringify(t)).digest("hex").slice(0,24)},uo=async(e,t)=>{const n=co(e);let o=e.contexts.get(n);if(o&&!o.released)return o.releaseTimer&&(clearTimeout(o.releaseTimer),o.releaseTimer=null,console.log(`[Context] Cancelled pending release for context "${n}"`)),o.refCount+=1,console.log(`[Context] Reusing existing context "${n}", refCount=${o.refCount}`),await o.ready,"function"==typeof t&&t(1),o;o?console.log(`[Context] Record exists but released=${o.released}, creating new context`):console.log(`[Context] No existing record for "${n}", creating new context`),o={key:n,refCount:1,ready:null,released:!1},e.contexts.set(n,o),o.ready=(async()=>{const r=Date.now(),a=await(async(e,t,n,o=null)=>{const{config:r,localPath:a,artifact:i}=e;if(e.localExists&&!t.has(a))return e.info.download.exists=!0,"function"==typeof n&&n(.5),a;if(r.model.local_path&&!r.model.allow_local_file)throw new Error("Local model path provided but `model.allow_local_file` is not enabled");const s=a;if(o){const t=o.getDownload(s);if(t){console.log(`[ensureModelFile] Waiting for global download: ${i.repoId}`);try{if(await t,await to(a,i.size))return e.localExists=!0,e.info.download.exists=!0,"function"==typeof n&&n(.5),a}catch(e){console.warn(`[ensureModelFile] Global download failed, will retry: ${e.message}`)}}}t.has(s)||t.set(s,(async()=>{if(i.isSplit&&i.splitCount>0){const e=/-(\d{5})-of-(\d{5})\.gguf$/,t=g.dirname(a),o=i.splitCount;let s=0;for(let a=1;a<=o;a+=1){const l=String(a).padStart(5,"0"),c=i.filename.replace(e,`-${l}-of-${String(o).padStart(5,"0")}.gguf`),u=`${r.model.base_url.replace(/\/+$/,"")}/${i.repoId}/resolve/${i.revision}/${c}`,d=g.join(t,c);await to(d)||await no(u,i.headers,d,null,e=>{if(e>=0&&Number.isFinite(e)){const t=(s+e)/o,r=Math.round(100*t);console.log(`Downloading model splits: ${Math.min(100,r)}%`),"function"==typeof n&&n(.5*t)}}),s+=1}}else console.log("Downloading model: 0%"),await no(i.url,i.headers,a,i.size,e=>{if(e>=0&&Number.isFinite(e)){const t=Math.round(100*e);console.log(`Downloading model: ${Math.min(100,t)}%`),"function"==typeof n&&n(.5*e)}});e.localExists=!0,e.info.download.exists=!0})());try{await t.get(s)}finally{t.delete(s)}return a})(e.plan,e.downloads,t,e.globalDownloadManager);"function"==typeof t&&t(.5);const i={model:a,n_threads:e.plan.info.runtime.n_threads,use_mmap:e.plan.info.runtime.use_mmap,use_mlock:e.plan.info.runtime.use_mlock,cpu_mask:e.plan.info.runtime.cpu_mask,cpu_strict:e.plan.info.runtime.cpu_strict,devices:e.plan.info.runtime.devices,n_ctx:e.plan.info.runtime.n_ctx,n_gpu_layers:e.plan.info.runtime.n_gpu_layers,n_parallel:e.plan.info.runtime.n_parallel,n_batch:e.plan.info.runtime.n_batch,flash_attn_type:e.plan.info.runtime.flash_attn_type,ctx_shift:e.plan.info.runtime.ctx_shift,kv_unified:e.plan.info.runtime.kv_unified,swa_full:e.plan.info.runtime.swa_full,lib_variant:e.plan.info.runtime.variant};let l;"off"!==e.plan.info.runtime.flash_attn_type&&(i.cache_type_k=e.plan.info.runtime.cache_type_k,i.cache_type_v=e.plan.info.runtime.cache_type_v),console.log("[Context] Load Options:",i);try{if(l=await s(i,e=>{"function"==typeof t&&(t(.5+.25*e),e%5==0&&console.log("[Context] Load Model Progress:",e))}),e.plan.info.runtime.n_parallel){if(!await l.parallel.enable({n_parallel:e.plan.info.runtime.n_parallel,n_batch:e.plan.info.runtime.n_batch}))throw new Error("Failed to enable parallel decoding mode for context")}return"function"==typeof t&&t(1),o.context=l,o.modelInfo=l.getModelInfo(),wn.addModelLoad({id:`${e.id}-${n}`,generatorId:e.id,contextKey:n,repoId:e.plan.info.model?.repoId||null,quantization:e.plan.info.model?.quantization||null,variant:e.plan.info.runtime?.variant||null,nCtx:e.plan.info.runtime?.n_ctx||null,nGpuLayers:e.plan.info.runtime?.n_gpu_layers||null,durationMs:Date.now()-r,success:!0}),o}catch(t){if(wn.addModelLoad({id:`${e.id}-${n}`,generatorId:e.id,contextKey:n,repoId:e.plan.info.model?.repoId||null,quantization:e.plan.info.model?.quantization||null,variant:e.plan.info.runtime?.variant||null,durationMs:Date.now()-r,success:!1,error:t?.message||String(t)}),l)try{l.release()}catch(e){}throw t}})();try{return await o.ready,o}catch(t){throw e.contexts.delete(n),t}},po=async(e,t,n=!1)=>{if(t.released)return!1;if(!n&&t.refCount>0)return!1;t.released=!0,e.contexts.delete(t.key);try{t.context?.parallel?.disable?.()}catch(e){}return await(t.context?.release?.()),!0},mo=async(e,t,n=!1)=>{if(t.releaseTimer&&(clearTimeout(t.releaseTimer),t.releaseTimer=null),n)t.refCount=0;else if(t.refCount=Math.max(0,t.refCount-1),t.refCount>0)return!1;const o=e.config.runtime.context_release_delay_ms;return n||o<=0?po(e,t):(console.log(`[Context] Scheduling release in ${o}ms for context "${t.key}"`),t.releaseTimer=setTimeout(async()=>{t.releaseTimer=null,t.refCount>0?console.log(`[Context] Release cancelled, refCount=${t.refCount} for context "${t.key}"`):(console.log(`[Context] Releasing context "${t.key}" after ${o}ms delay`),await po(e,t))},o),!0)};async function fo(e,t,n={}){const{globalDownloadManager:o=null}=n,r=An(t),a=await ao(r),i=new Wn(r,a);await i.initialize();const s={id:e,type:"ggml-llm",config:r,plan:a,info:a.info,contexts:new Map,downloads:new Map,globalDownloadManager:o,sessionCache:i,finalized:!1};return{id:e,type:"ggml-llm",info:a.info,contexts:s.contexts,initContext:async(e={})=>{const{onProgress:t}=e,n=await uo(s,t);return{modelInfo:n.modelInfo?{...n.modelInfo}:null,runtime:{...s.plan.info.runtime},download:{...s.plan.info.download}}},completion:async(e={})=>{const{options:t={},useCache:n=!0}=e,o=co(s),r=s.contexts.get(o);if(!r)throw new Error(`Context "${o}" not initialized`);await r.ready;let a=t.prompt||"";if(!a&&t.messages){const e=await r.context.getFormattedChat(t.messages,t.chat_template||t.chatTemplate,{jinja:t.jinja??!0,tools:t.tools,parallel_tool_calls:t.parallel_tool_calls,tool_choice:t.tool_choice,enable_thinking:t.enable_thinking,add_generation_prompt:t.add_generation_prompt,now:t.now,chat_template_kwargs:t.chat_template_kwargs});a=e?.prompt||e||""}if(n&&s.sessionCache.enabled&&a){const{options:e}=await s.sessionCache.prepareCompletionOptions(t,a),n=await s.sessionCache.generateTempStatePath(),o=await r.context.tokenize(a),i=o?.tokens?.length||0,l={...e,save_state_path:n,save_state_size:i};console.log(`[SessionCache] save_state_size=${i} (prompt tokens)`);const c={repoId:s.plan.info.model?.repoId||null,quantization:s.plan.info.model?.quantization||null,variant:s.plan.info.runtime?.variant||null};return lo(r.context,l,s.sessionCache,a,n,i,s.id,c)}const i={repoId:s.plan.info.model?.repoId||null,quantization:s.plan.info.model?.quantization||null,variant:s.plan.info.runtime?.variant||null};return so(r.context,t,s.id,i)},tokenize:async(e={})=>{const{text:t="",params:n={}}=e,o=co(s),r=s.contexts.get(o);if(!r)throw new Error(`Context "${o}" not initialized`);await r.ready;const a=await r.context.tokenize(t,n);if(!a)return{tokens:[]};const i=Array.from(a.tokens??[]).map(e=>Number(e));return{...a,tokens:i}},detokenize:async(e={})=>{const{tokens:t=[]}=e,n=co(s),o=s.contexts.get(n);if(!o)throw new Error(`Context "${n}" not initialized`);await o.ready;const r=t.map(e=>Number(e));return o.context.detokenize(r)},applyChatTemplate:async(e={})=>{const{messages:t=[],template:n,params:o}=e,r=co(s),a=s.contexts.get(r);if(!a)throw new Error(`Context "${r}" not initialized`);return await a.ready,await a.context.getFormattedChat(t,n,o)},releaseContext:async()=>{if(s.finalized)return!1;const e=co(s),t=s.contexts.get(e);return!!t&&mo(s,t,!1)},finalize:async()=>{if(s.finalized)return;s.finalized=!0;const e=Array.from(s.contexts.values()),t=e.map(e=>e.released||e.releaseTimer?Promise.resolve(!1):(e.refCount=Math.max(0,e.refCount-1),e.refCount>0?Promise.resolve(!1):po(s,e)));await Promise.allSettled(t);(0===e.length||e.every(e=>e.released))&&await s.sessionCache.cleanup()},getStatus:()=>{const e=[],t=Array.from(s.contexts.entries()).map(([t,n])=>{const o={key:t,refCount:n.refCount,hasModel:Boolean(n.context)},r=n.context.parallel.getStatus();return o.parallelStatus=r,e.push({contextKey:t,...r}),o});return{id:s.id,type:s.type,repoId:s.plan.info.model?.repoId||null,quantization:s.plan.info.model?.quantization||null,variant:s.plan.info.runtime?.variant||null,nCtx:s.plan.info.runtime?.n_ctx||null,nParallel:s.plan.info.runtime?.n_parallel||null,contexts:t,parallelStatuses:e}},subscribeParallelStatus:e=>{const t=Array.from(s.contexts.entries()).map(([t,n])=>n.context.parallel.subscribeToStatus(n=>{e({contextKey:t,...n})}));return{remove:()=>{t.forEach(e=>{e?.remove&&e.remove()})}}},hasPendingReleases:()=>Array.from(s.contexts.values()).some(e=>!e.released&&(e.releaseTimer||e.refCount>0)),resetFinalized:()=>{s.finalized=!1}}}const ho=e=>{const t=An(e);return t.model.repo_id||t.model.repository||t.model.model||null};async function yo(e,t,n={}){const{onProgress:o,onComplete:r,onError:a}=n;try{const n=An(e),i=await oo(n),s=eo(n,i),{repoId:l}=i;if(await to(s,i.size))return console.log(`[Download] Model already exists: ${l} at ${s}`),"function"==typeof r&&r({localPath:s,repoId:l,alreadyExists:!0}),{started:!1,localPath:s,repoId:l,alreadyExists:!0};const c=t.getDownload(s);if(c)return console.log(`[Download] Already downloading: ${l}`),c.then(()=>{"function"==typeof r&&r({localPath:s,repoId:l,joinedExisting:!0})}).catch(e=>{"function"==typeof a&&a(e)}),{started:!1,localPath:s,repoId:l,alreadyDownloading:!0};console.log(`[Download] Starting download: ${l}`);const u=(async()=>{try{if(i.isSplit&&i.splitCount>0){const e=/-(\d{5})-of-(\d{5})\.gguf$/,t=g.dirname(s),r=i.splitCount;let a=0;for(let s=1;s<=r;s+=1){const c=String(s).padStart(5,"0"),u=i.filename.replace(e,`-${c}-of-${String(r).padStart(5,"0")}.gguf`),d=`${n.model.base_url.replace(/\/+$/,"")}/${i.repoId}/resolve/${i.revision}/${u}`,p=g.join(t,u);await to(p)||await no(d,i.headers,p,null,e=>{if(e>=0&&Number.isFinite(e)){const t=(a+e)/r;console.log(`[Download] ${l}: ${Math.round(100*t)}%`),"function"==typeof o&&o(t)}}),a+=1}}else await no(i.url,i.headers,s,i.size,e=>{e>=0&&Number.isFinite(e)&&(console.log(`[Download] ${l}: ${Math.round(100*e)}%`),"function"==typeof o&&o(e))});console.log(`[Download] Completed: ${l}`),"function"==typeof r&&r({localPath:s,repoId:l})}catch(e){throw console.error(`[Download] Failed: ${l}`,e.message),"function"==typeof a&&a(e),e}finally{t.deleteDownload(s)}})();return t.setDownload(s,u),{started:!0,localPath:s,repoId:l}}catch(e){return console.error("[Download] Failed to start download:",e.message),"function"==typeof a&&a(e),{started:!1,localPath:null,repoId:null,error:e.message}}}const go=e=>e?"number"==typeof e.score&&Number.isFinite(e.score)?Number(e.score):Lt(e):0;async function bo(e=null,t={}){const{threshold:n=1.1,includeBreakdown:o=!1,config:r,...s}=t;let l=null,c=null,u=null,d=null,p=null,m=null;if(r)try{const{modelBytes:e,kvCacheBytes:t,limitedKvCacheBytes:n,memoryLimitedCtx:o,kvInfo:a,quantization:i}=await async function(e){const t=An(e),n=await oo(t),o=await Xn(n.url,n.headers,t.runtime.cache_dir),{arch:r,nCtxTrain:a,nLayer:i,nEmbd:s,nHead:l,nHeadKv:c,nEmbdHeadK:u,nEmbdHeadV:d,quantVersion:p,fileType:m}=cn(o),f=Number.isFinite(Number(i))?Number(i):0,h=Number.isFinite(Number(s))?Number(s):0,y=Number.isFinite(Number(l))?Number(l):0,g=Number.isFinite(Number(c))?Number(c):y,b=y>0&&h>0?h/y:128,w=null!=u&&Number.isFinite(Number(u))?Number(u):b,v=null!=d&&Number.isFinite(Number(d))?Number(d):b,_=ln({arch:r,metadata:o,nLayer:f}),x=_&&Number.isFinite(Number(_.kvLayers))?Number(_.kvLayers):f,S=Math.max(0,Math.floor(Number(x)||0)),C=(t.model.n_ctx?Number(t.model.n_ctx):null)||a||4096,k={k:t.model.cache_type_k,v:t.model.cache_type_v},T=n.size>0?n.size:0,M=un({layerCount:S,headKvCount:g,embdHeadKCount:w,embdHeadVCount:v,cacheTypes:k,swaConfig:_,kvUnified:t.model.kv_unified,nParallel:t.model.n_parallel,swaFull:t.model.swa_full}),E=null!=t.backend?.gpu_memory_fraction?Math.min(1,Math.max(0,Number(t.backend.gpu_memory_fraction))):qn.backend.gpu_memory_fraction||1,$=null!=t.backend?.cpu_memory_fraction?Math.min(1,Math.max(0,Number(t.backend.cpu_memory_fraction))):Pn,B=M(C),I=await ro(t,{modelBytes:T,kvCacheBytes:B}),O=(I.selected.totalMemory||0)*E,N=Math.max(0,Pt.totalmem()*$),P=I.selected.hasGpu?O:N,q=dn({maxCtx:C,availableMemory:P,modelBytes:T,kvBytesForCtx:M}),z=M(C),D=M(q);return{kvInfo:{nCtxTrain:a,nLayer:f,nEmbd:h,nHeadKv:g,nEmbdHeadK:w,nEmbdHeadV:v,nHeadCount:y,nHeadKvCount:g,kvLayerCount:S,swa:_?.enabled?{window:_.window,pattern:_.pattern,denseFirst:_.denseFirst,type:_.type,layers:_.swaLayers}:null},modelBytes:T,kvCacheBytes:z,limitedKvCacheBytes:D,memoryLimitedCtx:q,quantization:{name:n.quantization||null,fileType:m,version:p}}}(r);l=e,c=t,u=n,d=o,p=a,m=i}catch(e){}const f=null!=r?.backend?.gpu_memory_fraction?Math.min(1,Math.max(0,Number(r.backend.gpu_memory_fraction))):void 0,h=null!=r?.backend?.cpu_memory_fraction?Math.min(1,Math.max(0,Number(r.backend.cpu_memory_fraction))):void 0,y=await en({...s,platform:process.platform,totalMemoryInBytes:Pt.totalmem(),backend:"ggml-llm",includeBreakdown:o,gpuMemoryFraction:f,cpuMemoryFraction:h,dependencies:{getBackendDevicesInfo:a,isLibVariantAvailable:i},modelBytes:l,kvCacheBytes:c,limitedKvCacheBytes:u}),g=y.selected,b=go(g);g.modelBytes=l||null,g.kvCacheBytes=c||null,g.memoryLimitedCtx=d||null,g.limitedKvCacheBytes=u||null,g.kvInfo=p||null,g.quantization=m||null;let w=null,v=null;if(e){const t=go(e);v={...e,score:t};let o="buttress",r="buttress-higher-score";if(y.ok)if(t||0===t){const e=v.fit,a=v.limitedFit,i=g?.fit,s=g?.limitedFit,l=e?.fitsInGpu||e?.fitsInCpu||a?.fitsInGpu||a?.fitsInCpu,c=i?.fitsInGpu||i?.fitsInCpu||s?.fitsInGpu||s?.fitsInCpu;l&&!c?(o="local",r="client-fits-in-memory"):c&&!l?(o="buttress",r="buttress-fits-in-memory"):t>b*n?(o="local",r="client-better"):b>t*n?(o="buttress",r="buttress-better"):(o="either",r="comparable-scores")}else o="buttress",r="missing-client-score";else o="local",r="buttress-unavailable";w={buttressScore:b,clientScore:t,threshold:n,recommendation:o,reason:r}}y.ok||w||(w={buttressScore:b,clientScore:e?.score??null,threshold:n,recommendation:"local",reason:"buttress-unavailable"});let _=null;return r&&(_={repoId:r.model?.repo_id||null,quantization:r.model?.quantization||null,nCtx:r.model?.n_ctx||null,cacheKType:r.model?.cache_type_k||"f16",cacheVType:r.model?.cache_type_v||"f16"}),{type:"ggml-llm",timestamp:(new Date).toISOString(),buttress:y,client:v,comparison:w,modelConfig:_}}const{WritableStream:wo}="undefined"!=typeof globalThis&&globalThis.ReadableStream&&globalThis.WritableStream?{ReadableStream:globalThis.ReadableStream,WritableStream:globalThis.WritableStream}:fn,vo=(e={},t={})=>(Object.entries(t||{}).forEach(([t,n])=>{n&&"object"==typeof n&&!Array.isArray(n)?(e[t]&&"object"==typeof e[t]||(e[t]={}),vo(e[t],n)):e[t]=n}),e),_o="https://huggingface.co",xo="https://huggingface.co/api",So=g.join(Pt.homedir(),".buttress","models"),Co=["cuda","vulkan","default"],ko=["q8_0","q5_1","q5_0","q4_1","q4_0"],To="fp16",Mo={backend:{type:"ggml-stt",variant:null,variant_preference:Co,gpu_memory_fraction:.85,cpu_memory_fraction:.5},model:{repo_id:"BricksDisplay/whisper-ggml",revision:"main",filename:null,url:null,quantization:null,preferred_quantizations:["q8_0",To,"q5_1"],allow_local_file:!1,local_path:null,api_base:xo,base_url:_o,use_gpu:!0,use_flash_attn:"auto"},runtime:{cache_dir:So,prefer_variants:[],huggingface_token:process.env.HUGGINGFACE_TOKEN||null,http_headers:{},max_threads:null,context_release_delay_ms:1e4}},Eo=(e,t=[])=>e||0===e?Array.isArray(e)?e.filter(e=>null!=e):[e]:[...t],$o=e=>{if(!e)return null;const t=String(e).toLowerCase();return["cuda","vulkan","default"].includes(t)?t:null},Bo=(e={})=>{const t=JSON.parse(JSON.stringify(Mo));if(vo(t,e),t.backend.variant=$o(t.backend.variant),t.backend.variant_preference=Array.from(new Set(Eo(t.backend.variant_preference||Co).map($o).filter(Boolean))),0===t.backend.variant_preference.length&&(t.backend.variant_preference=[...Co]),t.runtime.prefer_variants=Array.from(new Set(Eo(t.runtime.prefer_variants).map($o).filter(Boolean))),t.model.preferred_quantizations=Array.from(new Set(Eo(t.model.preferred_quantizations||t.model.quantizations).map(e=>e?String(e).toLowerCase():null).filter(Boolean))),t.model.quantization){const e=String(t.model.quantization).toLowerCase();t.model.preferred_quantizations.includes(e)||t.model.preferred_quantizations.unshift(e)}return t.model.base_url=t.model.base_url||_o,t.model.api_base=t.model.api_base||xo,t.runtime.cache_dir=t.runtime.cache_dir?g.resolve(t.runtime.cache_dir):So,t.runtime.context_release_delay_ms=Math.max(0,Number(t.runtime.context_release_delay_ms)||Mo.runtime.context_release_delay_ms),t},Io=e=>{const t=e.toLowerCase();return ko.find(e=>t.includes(e))||null},Oo=async e=>{await(0,y.mkdir)(e,{recursive:!0})},No=(e,t,n=So)=>{const o=(0,qt.createHash)("sha256").update(e).digest("hex");return g.join(((e=So)=>g.join(e,".metadata-cache"))(n),t,`${o}.json`)},Po=async(e,t,n,o=So)=>{try{const r=No(e,t,o);await Oo(g.dirname(r)),await(0,y.writeFile)(r,JSON.stringify(n),"utf-8")}catch(e){}},qo=async(e,t={})=>{if("function"!=typeof fetch)throw new Error("Global fetch is not available in this runtime");const n=await fetch(e,{...t,method:"HEAD"});if(!n.ok)throw new Error(`Failed to fetch headers for ${e}: ${n.status} ${n.statusText}`);return n},zo=(e,t)=>{if(e.model.local_path)return g.resolve(e.model.local_path);const n=t.repoId.split("/"),o=g.join(e.runtime.cache_dir,...n,t.revision);return g.join(o,t.filename)},Do=async(e,t)=>{try{const n=await(0,y.stat)(e);return!t||n.size===t}catch(e){return!1}},Ao=async(e,t,n,o,r)=>{if("function"!=typeof fetch)throw new Error("Global fetch is not available in this runtime");await Oo(g.dirname(n));const a=await fetch(e,{headers:t});if(!a.ok||!a.body)throw new Error(`Failed to download ${e}: ${a.status} ${a.statusText}`);const i=await(0,y.open)(n,"w"),s=Number(a.headers.get("content-length"))||o||0;let l=0,c=.05;try{await a.body.pipeTo(new wo({async write(e){if(await i.write(e),l+=e.byteLength,"function"==typeof r&&s>0){const e=Math.min(1,l/s);for(;e>=c;)r(c),c+=.05}},async close(){await i.close(),"function"==typeof r&&r(1)},async abort(e){throw await i.close().catch(()=>{}),await(0,y.unlink)(n).catch(()=>{}),e}}))}catch(e){throw await i.close().catch(()=>{}),await(0,y.unlink)(n).catch(()=>{}),e}if(o){const e=await(0,y.stat)(n);if(e.size!==o)throw await(0,y.unlink)(n).catch(()=>{}),new Error(`Downloaded file size mismatch, expected ${o} got ${e.size}`)}},Fo=async e=>{const t=e.model.repo_id||e.model.repository||e.model.model;if(!t)throw new Error("`model.repo_id` is required in Buttress backend config");const n=e.model.revision||"main",o=e.runtime.cache_dir,r=JSON.stringify({repoId:t,revision:n,filename:e.model.filename,url:e.model.url,quantization:e.model.quantization,preferred_quantizations:e.model.preferred_quantizations}),a=await(async(e,t,n=So)=>{try{const o=No(e,t,n),r=await(0,y.readFile)(o,"utf-8");return JSON.parse(r)}catch(e){return null}})(r,"artifact-info",o);if(a)return a;const i={...e.runtime.http_headers||{}};if(e.runtime.huggingface_token&&(i.Authorization=`Bearer ${e.runtime.huggingface_token}`),e.model.url){const a=await qo(e.model.url,{headers:i}),s=Number(a.headers.get("content-length"))||null,l=e.model.filename||e.model.url.split("/").pop(),c={repoId:t,revision:n,filename:l,url:e.model.url,size:s,quantization:Io(l||""),headers:i};return await Po(r,"artifact-info",c,o),c}let{filename:s}=e.model,l=e.model.quantization&&String(e.model.quantization).toLowerCase();const c=await(async(e,t={})=>{if("function"!=typeof fetch)throw new Error("Global fetch is not available in this runtime");const n=await fetch(e,t);if(!n.ok){const t=await n.text().catch(()=>"");throw new Error(`Failed to fetch ${e}: ${n.status} ${n.statusText} ${t}`.trim())}return n.json()})(`${e.model.api_base}/models/${t}?revision=${n}&blobs=true`,{headers:i}),u=(c?.siblings||c?.files||[]).map(e=>e.rfilename||e.path||e.filename).filter(e=>"string"==typeof e&&e.endsWith(".bin"));if(0===u.length)throw new Error(`No model artifacts found in repo ${t}`);const d=e.model.preferred_quantizations.length>0?e.model.preferred_quantizations:ko;if(s)l||(l=Io(s));else{const e=(()=>{for(const e of d)if(e===To){const e=u.find(e=>{const t=e.toLowerCase();return!ko.some(e=>t.includes(e))});if(e)return{filename:e,quantization:null}}else{const t=u.find(t=>t.toLowerCase().includes(e));if(t)return{filename:t,quantization:e}}return null})()||{filename:u[0],quantization:null},{filename:t,quantization:n}=e;s=t,l=n||Io(s)}const p=`${e.model.base_url.replace(/\/+$/,"")}/${t}/resolve/${n}/${s}`,m=await qo(p,{headers:i}),f={repoId:t,revision:n,filename:s,url:p,size:Number(m.headers.get("content-length"))||null,quantization:l,headers:i,isSplit:!1,splitCount:0};return await Po(r,"artifact-info",f,o),f},Ro=async(e,{modelBytes:t=null,processingBytes:n=null}={})=>{const o=(e=>{const t=[];return e.backend.variant&&t.push(e.backend.variant),e.runtime.prefer_variants.length>0&&t.push(...e.runtime.prefer_variants),t.push(...e.backend.variant_preference),t.push("default"),Array.from(new Set(t.map($o).filter(Boolean)))})(e),[r,...s]=o,l=null!=e.backend?.gpu_memory_fraction?Math.min(1,Math.max(0,Number(e.backend.gpu_memory_fraction))):Mo.backend.gpu_memory_fraction||1,c=null!=e.backend?.cpu_memory_fraction?Math.min(1,Math.max(0,Number(e.backend.cpu_memory_fraction))):.5,u=await en({platform:process.platform,totalMemoryInBytes:Pt.totalmem(),backend:"ggml-stt",variant:r||null,preferVariants:s,variantPreference:e.backend.variant_preference,gpuMemoryFraction:l,cpuMemoryFraction:c,dependencies:{getBackendDevicesInfo:a,isLibVariantAvailable:i},modelBytes:t,kvCacheBytes:n}),d=e=>({...e,devices:Array.isArray(e.devices)?e.devices:[],ok:e.ok,hasGpu:Boolean(e.hasGpu),totalMemory:e.gpuTotalBytes||e.totalMemory||0,error:e.ok?null:new Error(e.error||`Variant ${e.variant} not available on this platform`)});if(!u.ok||!u.selected){const e=(u.attempts||[]).map(e=>`${e.variant}: ${e.error||"unknown error"}`).join("; ");throw new Error(`Unable to initialize any backend variant (${o.join(", ")}). Errors: ${e}`)}const p=(u.attempts||[]).map(d);return{selected:d(u.selected),attempts:p}},Uo=async e=>{const t=await Fo(e),n=on({modelBytes:t.size>0?t.size:0}),o=await Ro(e,{modelBytes:n.modelBytes,processingBytes:n.processingBufferBytes});let r=o.selected.hasGpu&&(void 0===o.selected.fit?.fitsInGpu||o.selected.fit.fitsInGpu);!1===e.model.use_gpu&&(r=!1);const a=e.model.use_flash_attn&&String(e.model.use_flash_attn).toLowerCase();let i;i="on"===a||"true"===a||"off"!==a&&"false"!==a&&r;const s=e.runtime.cache_dir,l=zo(e,t),c=await Do(l,t.size);return{config:e,info:{ok:!0,backend:"ggml-stt",model:{repoId:t.repoId,revision:t.revision,filename:t.filename,quantization:t.quantization,url:t.url,sizeBytes:t.size},runtime:{variant:o.selected.variant,use_gpu:r,use_flash_attn:i,max_threads:e.runtime.max_threads?Number(e.runtime.max_threads):null},resources:{...n,gpuCapacityBytes:o.selected.gpuTotalBytes,gpuUsableBytes:o.selected.gpuUsableBytes,cpuUsableBytes:o.selected.cpuUsableBytes,fit:o.selected.fit},devices:{selected:o.selected,attempts:o.attempts},download:{cacheDir:s,localPath:l,exists:c},timestamp:(new Date).toISOString()},artifact:t,memory:n,devices:o,localPath:l,localExists:c}};class jo{constructor(){this.queue=[],this.processing=!1,this.currentTaskId=null}async enqueue(e,t=null){return new Promise((n,o)=>{this.queue.push({task:e,resolve:n,reject:o,taskId:t}),this.processNext()})}async processNext(){if(this.processing||0===this.queue.length)return;this.processing=!0;const{task:e,resolve:t,reject:n,taskId:o}=this.queue.shift();this.currentTaskId=o;try{t(await e())}catch(e){n(e)}finally{this.processing=!1,this.currentTaskId=null,this.processNext()}}getStatus(){return{processing:this.processing,queuedCount:this.queue.length,currentTaskId:this.currentTaskId}}}const Lo=e=>{if(!e)return null;if(e instanceof ArrayBuffer)return e;if(ArrayBuffer.isView(e))return e.buffer;if("string"==typeof e){const t=e.startsWith("data:")?e.split(",")[1]||"":e,n=Buffer.from(t,"base64");return n.buffer.slice(n.byteOffset,n.byteOffset+n.byteLength)}throw new Error("Unsupported audioData format, expected base64 string or ArrayBuffer")},Go=async(e,t)=>{if(e.contextRecord&&!e.contextRecord.released)return e.contextRecord.releaseTimer&&(clearTimeout(e.contextRecord.releaseTimer),e.contextRecord.releaseTimer=null,console.log("[Context] Cancelled pending STT release")),e.contextRecord.refCount+=1,console.log(`[Context] Reusing existing STT context, refCount=${e.contextRecord.refCount}`),await e.contextRecord.ready,"function"==typeof t&&t(1),e.contextRecord;e.contextRecord?console.log(`[Context] STT record exists but released=${e.contextRecord.released}, creating new context`):console.log("[Context] No existing STT record, creating new context");const n={refCount:1,ready:null,released:!1};e.contextRecord=n,n.ready=(async()=>{const o=Date.now();try{const r=await(async(e,t,n,o=null)=>{const{localPath:r,artifact:a,config:i}=e;if(e.localExists)return"function"==typeof n&&n(1),r;if(o){const t=o.getDownload(r);if(t){console.log(`[ensureModelFile] Waiting for global STT download: ${a.repoId}`);try{if(await t,await Do(r,a.size))return e.localExists=!0,e.info.download.exists=!0,"function"==typeof n&&n(1),r}catch(e){console.warn(`[ensureModelFile] Global STT download failed, will retry: ${e.message}`)}}}const s=t.get(r);if(s)return await s,"function"==typeof n&&n(1),r;const l=(async()=>{if(i.model.allow_local_file){if(!await Do(r,a.size))throw new Error(`Local model file not found: ${r}`);return r}return await Ao(a.url,a.headers,r,a.size,n),r})();t.set(r,l);try{return await l,r}finally{t.delete(r)}})(e.plan,e.downloads,t,e.globalDownloadManager),a=await l({filePath:r,useFlashAttn:"on"===e.plan.info.runtime.flash_attn_type,useGpu:e.plan.info.runtime.n_gpu_layers>0,nThreads:e.plan.info.runtime.n_threads},e.plan.info.runtime.variant);n.context=a;try{n.modelInfo=await a.getModelInfo()}catch(e){n.modelInfo=null}return vn.addModelLoad({id:e.id,generatorId:e.id,repoId:e.plan.info.model?.repoId||null,quantization:e.plan.info.model?.quantization||null,variant:e.plan.info.runtime?.variant||null,useGpu:e.plan.info.runtime?.use_gpu||!1,durationMs:Date.now()-o,success:!0}),n}catch(t){throw vn.addModelLoad({id:e.id,generatorId:e.id,repoId:e.plan.info.model?.repoId||null,quantization:e.plan.info.model?.quantization||null,variant:e.plan.info.runtime?.variant||null,durationMs:Date.now()-o,success:!1,error:t?.message||String(t)}),t}})();try{return await n.ready,"function"==typeof t&&t(1),n}catch(t){throw e.contextRecord=null,t}},Ho=async(e,t,n=!1)=>!t.released&&(!(!n&&t.refCount>0)&&(t.released=!0,e.contextRecord=null,await(t.context?.release?.()),!0)),Jo=async(e,t,n=!1)=>{if(t.releaseTimer&&(clearTimeout(t.releaseTimer),t.releaseTimer=null),n)t.refCount=0;else if(t.refCount=Math.max(0,t.refCount-1),t.refCount>0)return!1;const o=e.config.runtime.context_release_delay_ms;return n||o<=0?Ho(e,t):(console.log(`[Context] Scheduling STT release in ${o}ms`),t.releaseTimer=setTimeout(async()=>{t.releaseTimer=null,t.refCount>0?console.log(`[Context] STT release cancelled, refCount=${t.refCount}`):(console.log(`[Context] Releasing STT context after ${o}ms delay`),await Ho(e,t))},o),!0)};async function Vo(e,t,n={}){const{globalDownloadManager:o=null}=n,r=Bo(t),a=await Uo(r),i={id:e,type:"ggml-stt",config:r,plan:a,info:a.info,contextRecord:null,downloads:new Map,globalDownloadManager:o,queue:new jo,finalized:!1},s=async(e={})=>{const{audioPath:t,audioData:n,options:o={}}=e,r=await Go(i),a={...o};i.plan.info.runtime.max_threads&&null==a.maxThreads&&(a.maxThreads=i.plan.info.runtime.max_threads);const s=`transcription-${Date.now()}-${Math.random().toString(36).slice(2,8)}`,l=Date.now();return i.queue.enqueue(async()=>{await r.ready;try{let e;if(n){const t=Lo(n),{promise:o}=r.context.transcribeData(t,a);e=await o}else{if(!t)throw new Error("audioPath or audioData is required for transcription");const n=g.resolve(t),{promise:o}=r.context.transcribe(n,a);e=await o}return vn.addTranscription({id:s,generatorId:i.id,repoId:i.plan.info.model?.repoId||null,quantization:i.plan.info.model?.quantization||null,variant:i.plan.info.runtime?.variant||null,durationMs:Date.now()-l,segmentCount:e?.segments?.length||0,textLength:e?.text?.length||0,success:!0}),e}catch(e){throw vn.addTranscription({id:s,generatorId:i.id,repoId:i.plan.info.model?.repoId||null,quantization:i.plan.info.model?.quantization||null,variant:i.plan.info.runtime?.variant||null,durationMs:Date.now()-l,success:!1,error:e?.message||String(e)}),e}},s)};return{id:e,type:"ggml-stt",info:a.info,queue:i.queue,initContext:async(e={})=>{const{onProgress:t}=e,n=await Go(i,t);return{modelInfo:n.modelInfo?{...n.modelInfo}:null,runtime:{...i.plan.info.runtime},download:{...i.plan.info.download}}},transcribe:async(e={})=>s(e),transcribeData:async(e={})=>s(e),releaseContext:async()=>{if(i.finalized)return!1;const e=i.contextRecord;return!!e&&Jo(i,e)},finalize:async()=>{if(i.finalized)return;i.finalized=!0;const e=i.contextRecord;e&&(e.released||e.releaseTimer||(e.refCount=Math.max(0,e.refCount-1),e.refCount>0||await Ho(i,e)))},getStatus:()=>({id:i.id,type:i.type,repoId:i.plan.info.model?.repoId||null,quantization:i.plan.info.model?.quantization||null,variant:i.plan.info.runtime?.variant||null,hasContext:Boolean(i.contextRecord?.context),contextRefCount:i.contextRecord?.refCount||0,queueStatus:i.queue.getStatus()}),hasPendingReleases:()=>{const e=i.contextRecord;return!!e&&(!e.released&&(e.releaseTimer||e.refCount>0))},resetFinalized:()=>{i.finalized=!1}}}const Ko=e=>{const t=Bo(e);return t.model.repo_id||t.model.repository||t.model.model||null};async function Yo(e,t,n={}){const{onProgress:o,onComplete:r,onError:a}=n;try{const n=Bo(e),i=await Fo(n),s=zo(n,i),{repoId:l}=i;if(await Do(s,i.size))return console.log(`[Download] STT model already exists: ${l} at ${s}`),"function"==typeof r&&r({localPath:s,repoId:l,alreadyExists:!0}),{started:!1,localPath:s,repoId:l,alreadyExists:!0};const c=t.getDownload(s);if(c)return console.log(`[Download] Already downloading STT model: ${l}`),c.then(()=>{"function"==typeof r&&r({localPath:s,repoId:l,joinedExisting:!0})}).catch(e=>{"function"==typeof a&&a(e)}),{started:!1,localPath:s,repoId:l,alreadyDownloading:!0};console.log(`[Download] Starting STT model download: ${l}`);const u=(async()=>{try{await Ao(i.url,i.headers,s,i.size,e=>{e>=0&&Number.isFinite(e)&&(console.log(`[Download] ${l}: ${Math.round(100*e)}%`),"function"==typeof o&&o(e))}),console.log(`[Download] Completed STT model: ${l}`),"function"==typeof r&&r({localPath:s,repoId:l})}catch(e){throw console.error(`[Download] Failed STT model: ${l}`,e.message),"function"==typeof a&&a(e),e}finally{t.deleteDownload(s)}})();return t.setDownload(s,u),{started:!0,localPath:s,repoId:l}}catch(e){return console.error("[Download] Failed to start STT download:",e.message),"function"==typeof a&&a(e),{started:!1,localPath:null,repoId:null,error:e.message}}}const Wo=e=>e?"number"==typeof e.score&&Number.isFinite(e.score)?Number(e.score):Lt(e):0;async function Qo(e,t=null,n={}){if("ggml-llm"===e)return bo(t,n);if("ggml-stt"===e)return async function(e=null,t={}){const{threshold:n=1.1,includeBreakdown:o=!1,config:r,...s}=t;let l=null,c=null,u=null;if(r)try{const e=Bo(r),t=await Fo(e);l=t.size??null,({processingBufferBytes:c}=on({modelBytes:l})),u=t.quantization||null}catch(e){}const d=null!=r?.backend?.gpu_memory_fraction?Math.min(1,Math.max(0,Number(r.backend.gpu_memory_fraction))):void 0,p=null!=r?.backend?.cpu_memory_fraction?Math.min(1,Math.max(0,Number(r.backend.cpu_memory_fraction))):void 0,m=await en({...s,platform:process.platform,totalMemoryInBytes:Pt.totalmem(),backend:"ggml-stt",includeBreakdown:o,gpuMemoryFraction:d,cpuMemoryFraction:p,dependencies:{getBackendDevicesInfo:a,isLibVariantAvailable:i},modelBytes:l,kvCacheBytes:c}),f=m.selected,h=Wo(f);f&&(f.modelBytes=l||null,f.processingBytes=c||null,f.quantization=u||null);let y=null,g=null;if(e){const t=Wo(e);g={...e,score:t};let o="buttress",r="buttress-higher-score";if(m.ok)if(t||0===t)if(e.fit&&f?.fit){const a=e.fit.fitsInGpu||e.fit.fitsInCpu,i=f.fit.fitsInGpu||f.fit.fitsInCpu;a&&!i?(o="local",r="client-fits-in-memory"):i&&!a?(o="buttress",r="buttress-fits-in-memory"):t>h*n?(o="local",r="client-better"):h>t*n?(o="buttress",r="buttress-better"):(o="either",r="comparable-scores")}else t>h*n?(o="local",r="client-better"):h>t*n?(o="buttress",r="buttress-better"):(o="either",r="comparable-scores");else o="buttress",r="missing-client-score";else o="local",r="buttress-unavailable";y={buttressScore:h,clientScore:t,threshold:n,recommendation:o,reason:r}}m.ok||y||(y={buttressScore:h,clientScore:e?.score??null,threshold:n,recommendation:"local",reason:"buttress-unavailable"});let b=null;return r&&(b={repoId:r.model?.repo_id||null,quantization:r.model?.quantization||null,filename:r.model?.filename||null}),{type:"ggml-stt",timestamp:(new Date).toISOString(),buttress:m,client:g,comparison:y,modelConfig:b}}(t,n);throw new Error(`Unknown backend type: ${e}`)}const Zo=t("node:fs"),Xo=t("node:child_process"),er=JSON.parse('{"UU":"@fugood/buttress-backend-core","rE":"2.23.0-beta.34"}'),tr=e=>{if(!e)return{repoId:null,filename:null};const[t,n]=e.split(":");return{repoId:t,filename:n||null}};async function nr({modelIds:e=[],defaultConfig:t=null}={}){const n=[];console.log(`${er.UU} v${er.rE}`),console.log("Generating model capabilities comparison...\n"),n.push(`${er.UU} v${er.rE}`),n.push("## Model Capabilities Comparison\n"),e&&0!==e.length||(console.error("Error: No model IDs provided"),process.exit(1));try{const o=(e={},t={})=>{const n=Array.isArray(e)?[...e]:{...e};return Object.entries(t||{}).forEach(([e,t])=>{t&&"object"==typeof t&&!Array.isArray(t)?n[e]=o(n[e]||{},t):n[e]=t}),n},r=t||{},{server:a,generators:i=[],...s}=r,l=e=>o(JSON.parse(JSON.stringify(s)),e||{}),u=e=>{if(Array.isArray(i)&&i.length>0){const t=i.filter(e=>"ggml-llm"===e?.type);if(t.length>0&&e){const n=t.find(t=>t.model?.repo_id===e);if(n)return l(n)}}return Object.keys(s).length>0?l({}):null},d=[];for(let t=0;t<e.length;t+=1){const n=e[t];console.log(`[${t+1}/${e.length}] Analyzing ${n}...`);let o=u(n);o={...o||{},model:{...s.runtime,...o?.model||{},repo_id:n}};const r=await Qo("ggml-llm",null,{config:o,includeBreakdown:!0});d.push({modelId:n,capabilities:r,modelInfo:r.buttress?.selected||null,modelConfig:r.modelConfig||null})}const p=e=>e?(e/1024/1024/1024).toFixed(2):"N/A",m=e=>e?"✅":"🚫";n.push("| Model ID | Quantization | Size (GB) | Context Size | KV Cache Size (GB) | Total Required Memory (GB) | Fits GPU (Full) | Fits CPU (Full) |"),n.push("|----------|--------------|-----------|--------------|--------------------|-----------------------------|-----------------|-----------------|"),d.forEach(({modelId:e,modelInfo:t,modelConfig:o})=>{const r=t?.quantization?.name?.toUpperCase()||"N/A",a=p(t?.modelBytes),i=o?.nCtx||t?.kvInfo?.nCtxTrain||"N/A",s=un(t),l=Number(i),c=t?.kvCacheBytes||(s&&Number.isFinite(l)&&l>0?s(l):s&&s(t?.kvInfo?.nCtxTrain||0))||null,u=p(c),d=p(t?.modelBytes&&c?t.modelBytes+c:t?.fit?.totalRequiredBytes),f=m(t?.fit?.fitsInGpu),h=m(t?.fit?.fitsInCpu);n.push(`| ${e} | ${r} | ${a} | ${i} | ${u} | ${d} | ${f} | ${h} |`);if((null!=t?.memoryLimitedCtx||null!=t?.limitedFit)&&(!t?.fit?.fitsInGpu||!t?.fit?.fitsInCpu)){const e=t?.memoryLimitedCtx||i,o=Number(e),r=t?.limitedKvCacheBytes||s&&Number.isFinite(o)&&o>0&&s(o)||null,l=p(r),c=p(t?.modelBytes&&r?t.modelBytes+r:t?.limitedFit?.totalRequiredBytes),f=m(t?.limitedFit?.fitsInGpu),h=m(t?.limitedFit?.fitsInCpu);(e!==i||l!==u||c!==d)&&n.push(`| ↳ Limited | - | ${a} | ${e} | ${l} | ${c} | ${f} | ${h} |`)}}),n.push("\n---"),n.push("\n### System Information");let f=null;if("win32"!==process.platform)try{f=(0,Xo.execSync)("uname -a",{encoding:"utf8"}).trim()}catch{}if(f?n.push(`- **System:** ${f}`):(n.push(`- **Hostname:** ${Pt.hostname()}`),n.push(`- **OS:** ${Pt.type()} ${Pt.release()}`)),n.push(`- **Platform:** ${process.platform}`),n.push(`- **CPU Cores:** ${Pt.cpus().length}`),n.push(`- **Total System Memory:** ${(Pt.totalmem()/1024/1024/1024).toFixed(2)} GB`),d.length>0){const e=d[0].capabilities,t=e.buttress?.selected;if(t){const e=t.cpuTotalBytes>0?(t.cpuUsableBytes/t.cpuTotalBytes*100).toFixed(0):0;if(n.push(`- **Usable CPU Memory:** ${(t.cpuUsableBytes/1024/1024/1024).toFixed(2)} GB (${e}% of ${(t.cpuTotalBytes/1024/1024/1024).toFixed(2)} GB)`),t.hasGpu){const e=t.devices.filter(e=>"gpu"===e.type);if(e.length>0){const o=e[0];n.push(`- **GPU Backend:** ${o.backend}`),n.push(`- **GPU Name:** ${o.deviceName}`),n.push(`- **GPU Total Memory:** ${(o.maxMemorySize/1024/1024/1024).toFixed(2)} GB`);const r=t.gpuTotalBytes>0?(t.gpuUsableBytes/t.gpuTotalBytes*100).toFixed(0):0;n.push(`- **GPU Usable Memory:** ${(t.gpuUsableBytes/1024/1024/1024).toFixed(2)} GB (${r}% of ${(t.gpuTotalBytes/1024/1024/1024).toFixed(2)} GB)`)}}else n.push("- **GPU:** Not available")}}n.push("\n### Command Used");const h=process.argv.slice(2).join(" ");if(n.push(`\`\`\`bash\n${process.argv[0]} ${process.argv[1]} ${h}\n\`\`\``),n.push("\n### Package Information"),n.push(`- **Name:** ${er.UU}`),n.push(`- **Version:** ${er.rE}`),er.description&&n.push(`- **Description:** ${er.description}`),t&&Object.keys(t).length>0){n.push("\n### Configuration"),n.push("<details>"),n.push("<summary>Click to expand TOML configuration</summary>"),n.push("\n```toml");try{const e=c.stringify(t);n.push(e)}catch(e){n.push("# Error serializing config"),n.push(JSON.stringify(t,null,2))}n.push("```"),n.push("</details>")}const y=`ggml-llm-model-capabilities-${(new Date).toISOString().replace(/[.:]/g,"-").split("T")[0]}.md`,b=g.join(process.cwd(),y);Zo.writeFileSync(b,n.join("\n"),"utf8"),console.log(`\nModel capabilities table saved to: ${b}`),process.exit(0)}catch(e){console.error("Failed to generate model table:",e.message),process.exit(1)}}async function or({modelId:e=null,defaultConfig:t=null}={}){console.log(`${er.UU} v${er.rE}`),console.log("Testing capabilities for backend: ggml-llm"),e&&console.log(`Model: ${e}`),console.log("--------------------------------");try{const n=t||{},{server:o,generators:r=[],...a}=n,i=(e={},t={})=>{const n=Array.isArray(e)?[...e]:{...e};return Object.entries(t||{}).forEach(([e,t])=>{t&&"object"==typeof t&&!Array.isArray(t)?n[e]=i(n[e]||{},t):n[e]=t}),n},s=e=>i(JSON.parse(JSON.stringify(a)),e||{});let l=(e=>{if(Array.isArray(r)&&r.length>0){const t=r.filter(e=>"ggml-llm"===e?.type);if(t.length>0&&e){const n=t.find(t=>t.model?.repo_id===e);if(n)return s(n)}}return Object.keys(a).length>0?s({}):null})(e);e&&(l={...l||{},model:{...l?.model||{},repo_id:e}});const c=await Qo("ggml-llm",null,{config:l,includeBreakdown:!0}),u=c.buttress?.selected||null,d=c.modelConfig||null;if(e||d?.repoId){console.log("\n=== Model Information ===");const t=e||d?.repoId;console.log(`Repository ID: ${t}`),d?.quantization&&console.log(`Quantization: ${d.quantization}`),d?.nCtx&&console.log(`Context Length: ${d.nCtx}`),u?.quantization?.name&&console.log(`Model Quantization: ${u.quantization.name.toUpperCase()}`);const n=d?.cache_type_k||"f16",o=d?.cache_type_v||"f16";if(console.log(`KV Cache Type: K=${n}, V=${o}`),u?.modelBytes&&u?.kvCacheBytes){if(console.log(`Model Size: ${(u.modelBytes/1024/1024/1024).toFixed(2)} GB`),u.kvInfo?console.log(`KV Cache Size: ${(u.kvCacheBytes/1024/1024/1024).toFixed(2)} GB (KV info: ${JSON.stringify(u.kvInfo)})`):console.log(`KV Cache Size: ${(u.kvCacheBytes/1024/1024/1024).toFixed(2)} GB`),console.log(`Total Required Memory: ${((u.modelBytes+u.kvCacheBytes)/1024/1024/1024).toFixed(2)} GB`),null!=u.memoryLimitedCtx){const e=u.memoryLimitedCtx,t=u.kvInfo?.nCtxTrain;t?console.log(`\nMemory-Limited Context: ${e} (Train: ${t})`):console.log(`\nMemory-Limited Context: ${e}`),null!=u.limitedKvCacheBytes&&console.log(`Limited KV Cache Size: ${(u.limitedKvCacheBytes/1024/1024/1024).toFixed(2)} GB`)}}else if(c.buttress?.selected?.fit){const{totalRequiredBytes:e}=c.buttress.selected.fit;console.log(`Total Required Memory: ${(e/1024/1024/1024).toFixed(2)} GB`)}}if(c.buttress?.selected){const{selected:e}=c.buttress;console.log("\n=== Hardware Information ===");let t=null;if("win32"!==process.platform)try{t=(0,Xo.execSync)("uname -a",{encoding:"utf8"}).trim()}catch{}t?console.log(`System: ${t}`):(console.log(`Hostname: ${Pt.hostname()}`),console.log(`OS: ${Pt.type()} ${Pt.release()}`)),console.log(`Platform: ${e.platform}`),console.log(`CPU Cores: ${Pt.cpus().length}`),console.log(`Total System Memory: ${(Pt.totalmem()/1024/1024/1024).toFixed(2)} GB`);const n=e.cpuTotalBytes>0?(e.cpuUsableBytes/e.cpuTotalBytes*100).toFixed(0):0;if(console.log(`Usable CPU Memory: ${(e.cpuUsableBytes/1024/1024/1024).toFixed(2)} GB (${n}% of ${(e.cpuTotalBytes/1024/1024/1024).toFixed(2)} GB)`),e.hasGpu){console.log("\n--- GPU Details ---");e.devices.filter(e=>"gpu"===e.type).forEach(t=>{console.log(`GPU Backend: ${t.backend}`),console.log(`GPU Name: ${t.deviceName}`),console.log(`GPU Total Memory: ${(t.maxMemorySize/1024/1024/1024).toFixed(2)} GB`);const n=e.gpuTotalBytes>0?(e.gpuUsableBytes/e.gpuTotalBytes*100).toFixed(0):0;console.log(`GPU Usable Memory: ${(e.gpuUsableBytes/1024/1024/1024).toFixed(2)} GB (${n}% of ${(e.gpuTotalBytes/1024/1024/1024).toFixed(2)} GB)`),t.metadata&&(t.metadata.hasBFloat16&&console.log("Supports BFloat16: Yes"),t.metadata.hasUnifiedMemory&&console.log("Unified Memory: Yes"))})}else console.log("GPU: Not available");console.log(`\nBackend Variant: ${e.variant}`),console.log(`Performance Score: ${e.score}`),e.fit&&(console.log("\n--- Model Fit Analysis ---"),console.log("Fits in GPU: "+(e.fit.fitsInGpu?"Yes":"No")),console.log("Fits in CPU: "+(e.fit.fitsInCpu?"Yes":"No")),console.log(`Limiting Factor: ${e.fit.limiting}`),e.limitedFit&&(console.log("\n--- Memory-Limited Fit Analysis ---"),console.log(`Limited Total Required: ${(e.limitedFit.totalRequiredBytes/1024/1024/1024).toFixed(2)} GB`),console.log("Fits in GPU (Limited): "+(e.limitedFit.fitsInGpu?"Yes":"No")),console.log("Fits in CPU (Limited): "+(e.limitedFit.fitsInCpu?"Yes":"No")),console.log(`Limiting Factor (Limited): ${e.limitedFit.limiting}`)))}console.log("\n=== Full Capabilities JSON ==="),console.log(JSON.stringify(c,null,2)),process.exit(0)}catch(e){console.error("Failed to get capabilities:",e.message),process.exit(1)}}async function rr({modelIds:e=[],defaultConfig:t=null}={}){const n=[];console.log(`${er.UU} v${er.rE}`),console.log("Generating STT model capabilities comparison...\n"),n.push(`${er.UU} v${er.rE}`),n.push("## STT Model Capabilities Comparison\n"),e&&0!==e.length||(console.error("Error: No model IDs provided"),process.exit(1));try{const o=(e={},t={})=>{const n=Array.isArray(e)?[...e]:{...e};return Object.entries(t||{}).forEach(([e,t])=>{t&&"object"==typeof t&&!Array.isArray(t)?n[e]=o(n[e]||{},t):n[e]=t}),n},r=t||{},{server:a,generators:i=[],...s}=r,l=e=>o(JSON.parse(JSON.stringify(s)),e||{}),u=e=>{if(Array.isArray(i)&&i.length>0){const t=i.filter(e=>"ggml-stt"===e?.type);if(t.length>0&&e){const n=t.find(t=>t.model?.repo_id===e);if(n)return l(n)}}return Object.keys(s).length>0?l({}):null},d=[];for(let t=0;t<e.length;t+=1){const n=e[t],{repoId:o,filename:r}=tr(n);console.log(`[${t+1}/${e.length}] Analyzing ${n}...`);let a=u(o);a={...a||{},model:{...a?.model||{},repo_id:o,...r&&{filename:r}}};const i=await Qo("ggml-stt",null,{config:a,includeBreakdown:!0});d.push({modelId:n,repoId:o,filename:r,capabilities:i,modelInfo:i.buttress?.selected||null,modelConfig:i.modelConfig||null})}const p=e=>e?(e/1024/1024).toFixed(1):"N/A",m=e=>e?"✅":"🚫";n.push("| Model | Size (MB) | Processing Buffer (MB) | Total Required (MB) | Fits GPU | Fits CPU |"),n.push("|-------|-----------|------------------------|---------------------|----------|----------|"),d.forEach(({modelId:e,modelInfo:t})=>{const o=p(t?.modelBytes),r=p(t?.processingBytes||t?.kvCacheBytes),a=p(t?.fit?.totalRequiredBytes),i=m(t?.fit?.fitsInGpu),s=m(t?.fit?.fitsInCpu);n.push(`| ${e} | ${o} | ${r} | ${a} | ${i} | ${s} |`)}),n.push("\n---"),n.push("\n### System Information");let f=null;if("win32"!==process.platform)try{f=(0,Xo.execSync)("uname -a",{encoding:"utf8"}).trim()}catch{}if(f?n.push(`- **System:** ${f}`):(n.push(`- **Hostname:** ${Pt.hostname()}`),n.push(`- **OS:** ${Pt.type()} ${Pt.release()}`)),n.push(`- **Platform:** ${process.platform}`),n.push(`- **CPU Cores:** ${Pt.cpus().length}`),n.push(`- **Total System Memory:** ${(Pt.totalmem()/1024/1024/1024).toFixed(2)} GB`),d.length>0){const e=d[0].capabilities,t=e.buttress?.selected;if(t){const e=t.cpuTotalBytes>0?(t.cpuUsableBytes/t.cpuTotalBytes*100).toFixed(0):0;if(n.push(`- **Usable CPU Memory:** ${(t.cpuUsableBytes/1024/1024/1024).toFixed(2)} GB (${e}% of ${(t.cpuTotalBytes/1024/1024/1024).toFixed(2)} GB)`),t.hasGpu){const e=t.devices.filter(e=>"gpu"===e.type);if(e.length>0){const o=e[0];n.push(`- **GPU Backend:** ${o.backend}`),n.push(`- **GPU Name:** ${o.deviceName}`),n.push(`- **GPU Total Memory:** ${(o.maxMemorySize/1024/1024/1024).toFixed(2)} GB`);const r=t.gpuTotalBytes>0?(t.gpuUsableBytes/t.gpuTotalBytes*100).toFixed(0):0;n.push(`- **GPU Usable Memory:** ${(t.gpuUsableBytes/1024/1024/1024).toFixed(2)} GB (${r}% of ${(t.gpuTotalBytes/1024/1024/1024).toFixed(2)} GB)`)}}else n.push("- **GPU:** Not available")}}n.push("\n### Command Used");const h=process.argv.slice(2).join(" ");if(n.push(`\`\`\`bash\n${process.argv[0]} ${process.argv[1]} ${h}\n\`\`\``),n.push("\n### Package Information"),n.push(`- **Name:** ${er.UU}`),n.push(`- **Version:** ${er.rE}`),er.description&&n.push(`- **Description:** ${er.description}`),t&&Object.keys(t).length>0){n.push("\n### Configuration"),n.push("<details>"),n.push("<summary>Click to expand TOML configuration</summary>"),n.push("\n```toml");try{const e=c.stringify(t);n.push(e)}catch(e){n.push("# Error serializing config"),n.push(JSON.stringify(t,null,2))}n.push("```"),n.push("</details>")}const y=`ggml-stt-model-capabilities-${(new Date).toISOString().replace(/[.:]/g,"-").split("T")[0]}.md`,b=g.join(process.cwd(),y);Zo.writeFileSync(b,n.join("\n"),"utf8"),console.log(`\nSTT model capabilities table saved to: ${b}`),process.exit(0)}catch(e){console.error("Failed to generate STT model table:",e.message),process.exit(1)}}async function ar({modelId:e=null,defaultConfig:t=null}={}){console.log(`${er.UU} v${er.rE}`),console.log("Testing capabilities for backend: ggml-stt"),e&&console.log(`Model: ${e}`),console.log("--------------------------------");try{const{repoId:n,filename:o}=tr(e),r=t||{},{server:a,generators:i=[],...s}=r,l=(e={},t={})=>{const n=Array.isArray(e)?[...e]:{...e};return Object.entries(t||{}).forEach(([e,t])=>{t&&"object"==typeof t&&!Array.isArray(t)?n[e]=l(n[e]||{},t):n[e]=t}),n},c=e=>l(JSON.parse(JSON.stringify(s)),e||{});let u=(e=>{if(Array.isArray(i)&&i.length>0){const t=i.filter(e=>"ggml-stt"===e?.type);if(t.length>0&&e){const n=t.find(t=>t.model?.repo_id===e);if(n)return c(n)}}return Object.keys(s).length>0?c({}):null})(n);n&&(u={...u||{},model:{...s.runtime,...u?.model||{},repo_id:n,...o&&{filename:o}}});const d=await Qo("ggml-stt",null,{config:u,includeBreakdown:!0}),p=d.buttress?.selected||null,m=d.modelConfig||null;if(n||m?.repoId){console.log("\n=== Model Information ===");const e=n||m?.repoId;console.log(`Repository ID: ${e}`),o&&console.log(`Filename: ${o}`),p?.modelBytes&&console.log(`Model Size: ${(p.modelBytes/1024/1024).toFixed(1)} MB`);const t=p?.processingBytes||p?.kvCacheBytes;if(t&&console.log(`Processing Buffer: ${(t/1024/1024).toFixed(1)} MB`),p?.modelBytes&&t)console.log(`Total Required Memory: ${((p.modelBytes+t)/1024/1024).toFixed(1)} MB`);else if(d.buttress?.selected?.fit){const{totalRequiredBytes:e}=d.buttress.selected.fit;console.log(`Total Required Memory: ${(e/1024/1024).toFixed(1)} MB`)}}if(d.buttress?.selected){const{selected:e}=d.buttress;console.log("\n=== Hardware Information ===");let t=null;if("win32"!==process.platform)try{t=(0,Xo.execSync)("uname -a",{encoding:"utf8"}).trim()}catch{}t?console.log(`System: ${t}`):(console.log(`Hostname: ${Pt.hostname()}`),console.log(`OS: ${Pt.type()} ${Pt.release()}`)),console.log(`Platform: ${e.platform}`),console.log(`CPU Cores: ${Pt.cpus().length}`),console.log(`Total System Memory: ${(Pt.totalmem()/1024/1024/1024).toFixed(2)} GB`);const n=e.cpuTotalBytes>0?(e.cpuUsableBytes/e.cpuTotalBytes*100).toFixed(0):0;if(console.log(`Usable CPU Memory: ${(e.cpuUsableBytes/1024/1024/1024).toFixed(2)} GB (${n}% of ${(e.cpuTotalBytes/1024/1024/1024).toFixed(2)} GB)`),e.hasGpu){console.log("\n--- GPU Details ---");e.devices.filter(e=>"gpu"===e.type).forEach(t=>{console.log(`GPU Backend: ${t.backend}`),console.log(`GPU Name: ${t.deviceName}`),console.log(`GPU Total Memory: ${(t.maxMemorySize/1024/1024/1024).toFixed(2)} GB`);const n=e.gpuTotalBytes>0?(e.gpuUsableBytes/e.gpuTotalBytes*100).toFixed(0):0;console.log(`GPU Usable Memory: ${(e.gpuUsableBytes/1024/1024/1024).toFixed(2)} GB (${n}% of ${(e.gpuTotalBytes/1024/1024/1024).toFixed(2)} GB)`),t.metadata&&(t.metadata.hasBFloat16&&console.log("Supports BFloat16: Yes"),t.metadata.hasUnifiedMemory&&console.log("Unified Memory: Yes"))})}else console.log("GPU: Not available");console.log(`\nBackend Variant: ${e.variant}`),console.log(`Performance Score: ${e.score}`),e.fit&&(console.log("\n--- Model Fit Analysis ---"),console.log("Fits in GPU: "+(e.fit.fitsInGpu?"Yes":"No")),console.log("Fits in CPU: "+(e.fit.fitsInCpu?"Yes":"No")),console.log(`Limiting Factor: ${e.fit.limiting}`))}console.log("\n=== Full Capabilities JSON ==="),console.log(JSON.stringify(d,null,2)),process.exit(0)}catch(e){console.error("Failed to get capabilities:",e.message),process.exit(1)}}const ir=new Map,sr={downloads:new Map,getDownload(e){return this.downloads.get(e)||null},setDownload(e,t){this.downloads.set(e,t)},deleteDownload(e){this.downloads.delete(e)},isDownloading(e){return this.downloads.has(e)},getActiveDownloads(){return Array.from(this.downloads.entries()).map(([e,t])=>({localPath:e,promise:t}))}},lr=(e,t)=>{const n=(e=>{const t=ir.get(e);if(!t)throw new Error(`Unknown generator id "${e}"`);return t})(e);if(n.type!==t)throw new Error(`Generator "${e}" does not support ${t} backend`);return n.instance};async function cr(e,t){const n={"ggml-llm":{create:fo,getId:ho},"ggml-stt":{create:Vo,getId:Ko}}[e];if(!n)throw new Error(`Unsupported backend type: ${e}`);const o=n.getId(t);if(!o)throw new Error("Buttress generator config missing repo identifier");const r=`${e}:${o}`,a=ir.get(r);if(a)return a.refCount+=1,a.instance.resetFinalized?.(),{id:a.id,info:a.instance.info};const i=await n.create(r,t,{globalDownloadManager:sr}),s={id:r,type:i.type,instance:i,refCount:1};return ir.set(r,s),{id:r,info:i.info}}async function ur(e){const t=ir.get(e);if(!t)return!1;if(t.refCount-=1,t.refCount<=0){await t.instance.finalize();(t.instance.hasPendingReleases?.()??!1)||ir.delete(e)}return!0}const dr={initContext:async(e,t)=>lr(e,"ggml-llm").initContext(t),completion:async(e,t)=>lr(e,"ggml-llm").completion(t),tokenize:async(e,t)=>lr(e,"ggml-llm").tokenize(t),detokenize:async(e,t)=>lr(e,"ggml-llm").detokenize(t),applyChatTemplate:async(e,t)=>lr(e,"ggml-llm").applyChatTemplate(t),releaseContext:async(e,t)=>lr(e,"ggml-llm").releaseContext(t)},pr={initContext:async(e,t)=>lr(e,"ggml-stt").initContext(t),transcribe:async(e,t)=>lr(e,"ggml-stt").transcribe(t),transcribeData:async(e,t)=>lr(e,"ggml-stt").transcribeData(t),releaseContext:async(e,t)=>lr(e,"ggml-stt").releaseContext(t)};function mr(e,t){return"ggml-llm"===e?ho(t):"ggml-stt"===e?Ko(t):null}const fr={getFullStatus:()=>function(e){return{timestamp:(new Date).toISOString(),ggmlLlm:Sn(e),ggmlStt:Cn(e)}}(ir),getGgmlLlmStatus:()=>Sn(ir),getGgmlSttStatus:()=>Cn(ir),subscribeToStatus:xn,subscribeToStatusWithId:function(e){return _n+=1,{subscriberId:_n,unsubscribe:xn(e)}},llmStatusTracker:wn,sttStatusTracker:vn,statusEmitter:yn};async function hr(e,t,n={}){const o={"ggml-llm":yo,"ggml-stt":Yo}[e];return o?o(t,sr,n):{started:!1,localPath:null,repoId:null,error:`Unknown backend type: ${e}`}}const{version:yr,name:gr}={version:"2.23.0-beta.34",name:"@fugood/buttress-server-poc"},br=async()=>{const e=`https://registry.npmjs.org/${gr}/latest`;try{const t=new AbortController,n=setTimeout(()=>t.abort(),3e3),o=await fetch(e,{headers:{Accept:"application/json"},signal:t.signal});if(clearTimeout(n),!o.ok)return null;return(await o.json()).version||null}catch{return null}},wr=(e,t)=>{if(!t)return!1;const n=e.split(/[.-]/),o=t.split(/[.-]/);for(let e=0;e<Math.max(n.length,o.length);e+=1){const t=parseInt(n[e])||0,r=parseInt(o[e])||0;if(r>t)return!0;if(r<t)return!1}return!1},vr=async()=>{try{const e=await br();e&&wr(yr,e)&&(e=>{console.log(""),console.log("╭─────────────────────────────────────────────────╮"),console.log("│ Update available! %s → %s",yr.padEnd(12),e.padEnd(12),"│"),console.log("│ │"),console.log("│ Run to upgrade: │"),console.log("│ npm install -g %s │",gr.padEnd(27)),console.log("╰─────────────────────────────────────────────────╯"),console.log("")})(e)}catch(e){}},_r=n.create(),xr=e=>{const{status:t}=e;return _r.router({getStatus:_r.procedure.query(()=>(console.log("[Status] Get full status"),t.getFullStatus())),getLlmStatus:_r.procedure.query(()=>(console.log("[Status] Get LLM status"),t.getGgmlLlmStatus())),getSttStatus:_r.procedure.query(()=>(console.log("[Status] Get STT status"),t.getGgmlSttStatus())),subscribe:_r.procedure.subscription(()=>te(n=>{console.log("[Status] Client subscribed to status changes");const{generatorRegistry:o}=e,r=[],a=(e,o)=>{if("ggml-llm"===o.type&&o.instance?.subscribeParallelStatus){const a=o.instance.subscribeParallelStatus(o=>{n.next({type:"parallelStatus",generatorId:e,data:o,fullStatus:t.getFullStatus()})});r.push({id:e,sub:a})}};Array.from(o.entries()).forEach(([e,t])=>{a(e,t)}),n.next({type:"initial",data:t.getFullStatus()});const i=t.subscribeToStatus(e=>{if("modelLoad"===e.type&&e.entry?.generatorId){const t=o.get(e.entry.generatorId);t&&a(e.entry.generatorId,t)}n.next({type:"change",data:e,fullStatus:t.getFullStatus()})});return()=>{console.log("[Status] Client unsubscribed from status changes"),r.forEach(({sub:e})=>e?.remove?.()),i()}}))})},Sr=new TextDecoder,Cr=(e,t)=>{if(!t||"*"===t)return"*";const n=e?.headers?.origin||"";return(Array.isArray(t)?t:[t]).includes(n)?n:""},kr=(e,t,n,o,r)=>{const a=Cr(e,r),i={"Content-Type":"application/json"};a&&(i["Access-Control-Allow-Origin"]=a),t.writeHead(n,i),t.end(JSON.stringify(o))},Tr=(e,t)=>{e.write(`data: ${JSON.stringify(t)}\n\n`)},Mr=(e,t)=>{const n=e.split("\n\n"),o=n.pop()||"";return n.forEach(e=>{if(!e.trim())return;const n=e.match(/^event:\s*(.+)$/m),o=e.match(/^data:\s*(.+)$/m);if(!n||!o)return;const r=n[1];let a;try{a=JSON.parse(o[1])}catch{return}t(r,a)}),o},Er=({backend:e,defaultConfig:t})=>{const n=t?.openai_compat?.cors_allowed_origins,o=new Map,r=async(e,t)=>{const n=e.getReader();let o="";try{let e=!1;for(;!e;){const{value:r,done:a}=await n.read();if(e=a,e)break;o+=Sr.decode(r,{stream:!0}),o=Mr(o,(e,n)=>{"token"===e&&t.onToken?t.onToken(n):"result"===e&&t.onResult?t.onResult(n):"error"===e&&t.onError&&t.onError(n)})}}finally{n.cancel().catch(()=>{})}},a=async(a,i)=>{let s;try{s=await(e=>new Promise((t,n)=>{let o="";e.on("data",e=>{o+=e.toString()}),e.on("end",()=>{try{t(o?JSON.parse(o):{})}catch(e){n(new Error("Invalid JSON body"))}}),e.on("error",n)}))(a)}catch(e){return void kr(a,i,400,{error:{message:e.message,type:"invalid_request_error"}},n)}const{messages:l=[],stream:c=!1,model:u,tools:d,temperature:p,stop:m}=s,f=s.top_p,h=s.max_tokens,y=s.presence_penalty,g=s.frequency_penalty,b=s.tool_choice,w=s.stream_options;if(l&&0!==l.length)try{const s=await(async n=>{const r=(t?.generators||[]).filter(e=>"ggml-llm"===e.type);if(0===r.length)throw new Error('No ggml-llm generator configured. Add a [[generators]] with type = "ggml-llm" to your config.');let a=r[0],i=n||a.model?.repo_id;if(n){const e=r.find(e=>e.model?.repo_id===n);e&&(a=e)}else i=a.model?.repo_id;const s=i,l=o.get(s);if(l?.initialized)return l;const{generators:c,server:u,...d}=t||{},p={...d,...a,model:{...a.model,repo_id:i}};console.log(`[OpenAI] Creating generator for ${s}`);const{id:m}=await e.startGenerator("ggml-llm",p),f={id:m,config:p,repoId:i,initialized:!1};return o.set(s,f),await e.ggmlLlm.initContext(m,{}),f.initialized=!0,console.log(`[OpenAI] Generator ready: ${s}`),f})(u),v=`chatcmpl-${Date.now()}-${Math.random().toString(36).slice(2,9)}`,_=Math.floor(Date.now()/1e3),x=s.repoId||"ggml-llm",S={reasoning_format:"auto",messages:l,jinja:!0,add_generation_prompt:!0};null!=p&&(S.temperature=p),null!=f&&(S.top_p=f),null!=h&&(S.n_predict=h),null!=m&&(S.stop=Array.isArray(m)?m:[m]),null!=y&&(S.presence_penalty=y),null!=g&&(S.frequency_penalty=g),null!=d&&(S.tools=d),null!=b&&(S.tool_choice=b);const C=await e.ggmlLlm.completion(s.id,{options:S});if(c){const e=Cr(a,n),t={"Content-Type":"text/event-stream","Cache-Control":"no-cache",Connection:"keep-alive"};e&&(t["Access-Control-Allow-Origin"]=e),i.writeHead(200,t);const o=!0===w?.include_usage;let s="",l="";const c=new Map,u=new Map;await r(C,{onToken:e=>{const t={};if(null!=e.content){const n=e.content;n.length>s.length&&(t.content=n.slice(s.length),s=n)}if(null!=e.reasoning_content){const n=e.reasoning_content;n.length>l.length&&(t.reasoning_content=n.slice(l.length),l=n)}if(null!=e.tool_calls&&e.tool_calls.length>0){const n=[];e.tool_calls.forEach((e,t)=>{const o={index:t};u.has(t)||(u.set(t,e.id||`call_${v}_${t}`),o.id=u.get(t),o.type="function");const r=e.function?.arguments||"",a=c.get(t)||"",i={};!c.has(t)&&e.function?.name&&(i.name=e.function.name),r.length>a.length&&(i.arguments=r.slice(a.length),c.set(t,r)),Object.keys(i).length>0?(o.function=i,n.push(o)):o.id&&(o.function={name:e.function?.name||"",arguments:""},n.push(o))}),n.length>0&&(t.tool_calls=n)}Object.keys(t).length>0&&Tr(i,{id:v,object:"chat.completion.chunk",created:_,model:x,choices:[{index:0,delta:t,finish_reason:null}]})},onResult:e=>{const t={id:v,object:"chat.completion.chunk",created:_,model:x,choices:[{index:0,delta:{},finish_reason:e.interrupted?"length":"stop"}]};o&&(t.usage={prompt_tokens:e.prompt_tokens||e.promptTokens||0,completion_tokens:e.tokens_predicted||e.tokensPredicted||0,total_tokens:(e.prompt_tokens||e.promptTokens||0)+(e.tokens_predicted||e.tokensPredicted||0)}),Tr(i,t)},onError:e=>{Tr(i,{error:{message:e.message,type:"server_error"}})}}),i.write("data: [DONE]\n\n"),i.end()}else{let e="",t=null,o=null,s="stop",l={prompt_tokens:0,completion_tokens:0,total_tokens:0};await r(C,{onToken:t=>{null!=t.content?e+=t.content:null!=t.token&&(e+=t.token)},onResult:n=>{n.text?e=n.text:n.content&&(e=n.content),n.reasoning_content&&(t=n.reasoning_content),n.tool_calls&&n.tool_calls.length>0?(o=n.tool_calls.map((e,t)=>({id:e.id||`call_${v}_${t}`,type:"function",function:{name:e.function?.name||"",arguments:e.function?.arguments||""}})),s="tool_calls"):s=n.interrupted?"length":"stop",l={prompt_tokens:n.prompt_tokens||n.promptTokens||0,completion_tokens:n.tokens_predicted||n.tokensPredicted||0,total_tokens:(n.prompt_tokens||n.promptTokens||0)+(n.tokens_predicted||n.tokensPredicted||0)}},onError:e=>{throw new Error(e.message)}});const c={role:"assistant",content:e||null};t&&(c.reasoning_content=t),o&&(c.tool_calls=o),kr(a,i,200,{id:v,object:"chat.completion",created:_,model:x,choices:[{index:0,message:c,finish_reason:s}],usage:l},n)}}catch(e){console.error("[OpenAI] Chat completion error:",e),kr(a,i,500,{error:{message:e.message||"Internal server error",type:"server_error"}},n)}else kr(a,i,400,{error:{message:"messages is required and must not be empty",type:"invalid_request_error"}},n)};return async(e,o,r)=>{if("OPTIONS"===e.method){const t=Cr(e,n),r={"Access-Control-Allow-Methods":"GET, POST, OPTIONS","Access-Control-Allow-Headers":"Content-Type, Authorization","Access-Control-Max-Age":"86400"};return t&&(r["Access-Control-Allow-Origin"]=t),o.writeHead(204,r),o.end(),!0}const{pathname:i}=r;return"/oai-compat/v1/models"===i&&"GET"===e.method?(await(async(e,o)=>{const r=(t?.generators||[]).filter(e=>"ggml-llm"===e.type).map(e=>({id:e.model?.repo_id||"ggml-llm",object:"model",created:Math.floor(Date.now()/1e3),owned_by:"local"}));0===r.length&&r.push({id:"ggml-llm",object:"model",created:Math.floor(Date.now()/1e3),owned_by:"local"}),kr(e,o,200,{object:"list",data:r},n)})(e,o),!0):"/oai-compat/v1/chat/completions"===i&&"POST"===e.method&&(await a(e,o),!0)}},$r=g.dirname((0,b.fileURLToPath)(import.meta.url)),Br=new TextDecoder,Ir=n.context().create({sse:{ping:{enabled:!0,intervalMs:3e4},maxDurationMs:18e5,client:{reconnectAfterInactivityMs:3e5}}}),Or=o.record(o.any()).optional(),Nr=o.object({id:o.string()}),Pr=o.object({id:o.string(),property:o.any().optional()}),qr=new Map,zr=o.object({type:o.string(),config:Or}),Dr=o.object({type:o.string().optional().default("ggml-llm"),config:Or,currentClientCapabilities:o.record(o.any()).optional(),options:o.record(o.any()).optional()}).optional(),Ar=(e={},t={})=>{const n=Array.isArray(e)?[...e]:{...e};return Object.entries(t||{}).forEach(([e,t])=>{t&&"object"==typeof t&&!Array.isArray(t)?n[e]=Ar(n[e]||{},t):n[e]=t}),n},Fr=e=>e&&"object"==typeof e?JSON.parse(JSON.stringify(e)):null,Rr=(e,t)=>{const n=Fr(e)||{},o=Fr(t)||{};return Ar(n,o)},Ur=(e=h,{defaultConfig:t}={})=>{const n=Fr(t)||{},r=Array.isArray(n.generators)?n.generators:[],{server:a,generators:i,...s}=n,l=e=>Ar(JSON.parse(JSON.stringify(s)),e||{}),c=(e,t)=>{if(r.length>0){const n=r.filter(t=>t?.type===e);if(n.length>0&&t){const e=n.find(e=>e.model?.repo_id===t);if(e)return l(e)}}return Object.keys(s).length>0?l({}):null};return Ir.router({getCapabilities:Ir.procedure.input(Dr).query(async({input:t})=>{console.log("[Server] Get Capabilities:",t);const n=t||{type:"ggml-llm"},{type:o="ggml-llm",config:r,currentClientCapabilities:a=null,options:i={}}=n,s=Fr(r),l=e.getModelIdentifier(o,s),u=c(o,l),d=Rr(u,r);if(0===Object.keys(d).length)throw new Error("Buttress server missing generator configuration");return d.backend=d.backend||{},d.backend.type||(d.backend.type=o),e.getCapabilities(o,a,{...i,config:d})}),startGenerator:Ir.procedure.input(zr).mutation(async({input:t})=>{console.log("[Server] Start Generator:",t);const{type:n,config:o}=t,r=Fr(o),a=e.getModelIdentifier(n,r),i=c(n,a),s=Rr(i,o);if(0===Object.keys(s).length)throw new Error("Buttress server missing generator configuration");return s.backend=s.backend||{},s.backend.type||(s.backend.type=n),e.startGenerator(n,s)}),finalizeGenerator:Ir.procedure.input(Nr).mutation(async({input:t})=>(console.log("[Server] Finalize Generator:",t),e.finalizeGenerator(t.id))),ggmlLlm:Ir.router({initContext:Ir.procedure.input(Pr).subscription(({input:t})=>te(n=>{console.log("[Server] Init Context:",t);let o=!0;return(async()=>{try{const r={...t.property,onProgress:e=>{o&&n.next({progress:e})}};await new Promise(e=>setTimeout(e));const a=await e.ggmlLlm.initContext(t.id,r);if(o){const{download:e,...t}=a||{};n.next({result:t}),n.complete()}}catch(e){o&&n.error(e)}})(),()=>{o=!1}})),createCompletion:Ir.procedure.input(Pr).mutation(async({input:t})=>{console.log("[Server] createCompletion:",t);const n=`completion-${Date.now()}-${Math.random().toString(36).slice(2,9)}`;try{const o=await e.ggmlLlm.completion(t.id,t.property);if(!o)throw new Error("Failed to create completion stream");return qr.set(n,{stream:o,cancel:null,generatorId:t.id}),{completionId:n}}catch(e){throw console.error("[Server] createCompletion error:",e),e}}),completion:Ir.procedure.input(Pr).subscription(({input:e})=>te(t=>{const n=e.property;console.log("[Server] completion subscription:",n);const o=qr.get(n);if(!o)return t.error(new Error(`Unknown completion ID: ${n}`)),()=>{};const{stream:r}=o,a=((e,t)=>{const n=e.getReader();let o=!0;return(async()=>{try{for(;o;){const{value:e,done:r}=await n.read();if(r||!o)break;if(e){const n=Br.decode(e,{stream:!0});t.next(n)}}t.complete()}catch(e){t.error(e)}finally{n.cancel().catch(()=>{})}})(),()=>{o=!1,n.cancel().catch(()=>{})}})(r,t);return o.cancel=a,()=>{console.log("[Server] completion subscription closed:",n),a(),qr.delete(n)}})),cancelCompletion:Ir.procedure.input(Pr).mutation(async({input:e})=>{const t=e.property;console.log("[Server] cancelCompletion:",t);const n=qr.get(t);return n?(n.cancel&&n.cancel(),qr.delete(t),{success:!0}):{success:!1,error:"Unknown completion ID"}}),tokenize:Ir.procedure.input(Pr).mutation(async({input:t})=>(console.log("[Server] Tokenize:",t),e.ggmlLlm.tokenize(t.id,t.property))),detokenize:Ir.procedure.input(Pr).mutation(async({input:t})=>(console.log("[Server] Detokenize:",t),e.ggmlLlm.detokenize(t.id,t.property))),applyChatTemplate:Ir.procedure.input(Pr).mutation(async({input:t})=>(console.log("[Server] Apply Chat Template:",t),e.ggmlLlm.applyChatTemplate(t.id,t.property))),releaseContext:Ir.procedure.input(Pr).mutation(async({input:t})=>(console.log("[Server] Release Context:",t),e.ggmlLlm.releaseContext(t.id,t.property)))}),ggmlStt:Ir.router({initContext:Ir.procedure.input(Pr).subscription(({input:t})=>te(n=>{console.log("[Server] Init STT Context:",t);let o=!0;return(async()=>{try{const r={...t.property,onProgress:e=>{o&&n.next({progress:e})}};await new Promise(e=>setTimeout(e));const a=await e.ggmlStt.initContext(t.id,r);if(o){const{download:e,...t}=a||{};n.next({result:t}),n.complete()}}catch(e){o&&n.error(e)}})(),()=>{o=!1}})),transcribeData:Ir.procedure.input(o.instanceof(FormData)).mutation(async({input:t})=>{const n=t,o=n.get("id"),r=n.get("options"),a=n.get("audioData"),i=r?JSON.parse(r):{};let s;if(a instanceof Blob||a&&"function"==typeof a.arrayBuffer){const e=await a.arrayBuffer();s=Buffer.from(e)}else s=Buffer.isBuffer(a)?a:a instanceof ArrayBuffer?Buffer.from(a):a;return console.log("[Server] Transcribe Data:",{id:o,audioDataLength:s?.length||0,options:i}),e.ggmlStt.transcribeData(o,{audioData:s,options:i})}),releaseContext:Ir.procedure.input(Pr).mutation(async({input:t})=>(console.log("[Server] Release STT Context:",t),e.ggmlStt.releaseContext(t.id,t.property)))}),status:xr(e)})},jr=({backend:e=h,router:t,createContext:n=()=>({}),defaultConfig:o,maxBodySize:r=104857600,enableOpenAICompat:a="1"===process.env.ENABLE_OPENAI_COMPAT_ENDPOINT}={})=>{const i=t||Ur(e,{defaultConfig:o}),s=a?Er({backend:e,defaultConfig:o}):null;return{server:Nt({router:i,createContext:n,basePath:"/trpc/",maxBodySize:r,middleware:async(e,t,n)=>{const o=e.headers["content-length"],r=e.headers["content-type"];console.log(`[Request] ${e.method} ${e.url} content-length=${o} content-type=${r}`);const a=new URL(e.url,`http://${e.headers.host}`);if("/status"!==a.pathname&&"/status/"!==a.pathname){if(s&&a.pathname.startsWith("/oai-compat/"))try{if(await s(e,t,a))return}catch(e){return console.error("[OpenAI] Handler error:",e),t.writeHead(500,{"Content-Type":"application/json"}),void t.end(JSON.stringify({error:{message:e.message,type:"server_error"}}))}n()}else(async e=>{try{const t=g.join($r,"..","public","status.html"),n=await(0,y.readFile)(t,"utf-8");return e.writeHead(200,{"Content-Type":"text/html; charset=utf-8"}),e.end(n),!0}catch(t){return e.writeHead(404,{"Content-Type":"text/plain"}),e.end("Status page not found"),!0}})(t)}}),router:i,backend:e,openaiEnabled:a}},Lr=async({port:e=2080,backend:t,router:n,createContext:o,defaultConfig:r,maxBodySize:a,enableOpenAICompat:i}={})=>{const{server:s,router:l,backend:c,openaiEnabled:u}=jr({backend:t,router:n,createContext:o,defaultConfig:r,maxBodySize:a,enableOpenAICompat:i});return s.server&&(void 0!==s.server.headersTimeout&&(s.server.headersTimeout=0),void 0!==s.server.requestTimeout&&(s.server.requestTimeout=0),void 0!==s.server.timeout&&(s.server.timeout=0)),await new Promise(t=>s.listen(e,t)),console.log(`[Server] Status page available at http://localhost:${e}/status`),{server:s,router:l,port:e,backend:c,openaiEnabled:u}},{testGgmlLlmCapabilities:Gr,testGgmlSttCapabilities:Hr,showModelsTable:Jr,showSttModelsTable:Vr,startModelDownload:Kr,globalDownloadManager:Yr}=h;export{vr as checkAndNotifyUpdates,br as checkForUpdates,wr as compareVersions,Ur as createRouter,jr as createServer,yr as currentVersion,Yr as globalDownloadManager,Jr as showModelsTable,Vr as showSttModelsTable,Kr as startModelDownload,Lr as startServer,Gr as testGgmlLlmCapabilities,Hr as testGgmlSttCapabilities};
package/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "@fugood/buttress-server-poc",
3
- "version": "2.23.0-beta.32",
3
+ "version": "2.23.0-beta.34",
4
4
  "type": "module",
5
5
  "main": "lib/index.js",
6
6
  "files": [
@@ -29,5 +29,5 @@
29
29
  "devDependencies": {
30
30
  "terser-webpack-plugin": "^5.0.3"
31
31
  },
32
- "gitHead": "46ce7c41468b88fec1d9428e27c7d9a7b8613ab5"
32
+ "gitHead": "dd12cf3ff4b81186f22628ab1116f144f73bbec6"
33
33
  }
@@ -513,12 +513,15 @@
513
513
  <th>Repo</th>
514
514
  <th>Quantization</th>
515
515
  <th>Variant</th>
516
- <th>Context</th>
516
+ <th>Context Size</th>
517
517
  <th>Parallel</th>
518
+ <th>Status</th>
518
519
  </tr>
519
520
  </thead>
520
521
  <tbody>
521
- ${generators.map(g => `
522
+ ${generators.map(g => {
523
+ const hasActiveContext = g.contexts?.some(c => c.hasModel) || false
524
+ return `
522
525
  <tr>
523
526
  <td class="mono">${escapeHtml(g.id?.slice(0, 20) || '-')}...</td>
524
527
  <td>${escapeHtml(g.repoId || '-')}</td>
@@ -526,8 +529,9 @@
526
529
  <td>${escapeHtml(g.variant || '-')}</td>
527
530
  <td>${g.nCtx?.toLocaleString() || '-'}</td>
528
531
  <td>${g.nParallel || '-'}</td>
532
+ <td>${hasActiveContext ? '<span class="badge badge-success">Active</span>' : '<span class="badge badge-warning">Inactive</span>'}</td>
529
533
  </tr>
530
- `).join('')}
534
+ `}).join('')}
531
535
  </tbody>
532
536
  </table>
533
537
  </div>
@@ -729,6 +733,8 @@
729
733
 
730
734
  renderHistory('llmCompletionHistory', llm.history?.completions || [], [
731
735
  { label: 'Time', render: i => `<span class="timestamp">${formatRelativeTime(i.timestamp)}</span>` },
736
+ { label: 'Model', render: i => escapeHtml(i.repoId || '-') },
737
+ { label: 'Quant', render: i => escapeHtml(i.quantization || '-') },
732
738
  { label: 'Tokens', render: i => `${i.tokensGenerated || 0}` },
733
739
  { label: 'Speed', render: i => `${i.tokensPerSecond?.toFixed(1) || '-'} t/s` },
734
740
  { label: 'Duration', render: i => `${(i.durationMs / 1000).toFixed(1)}s` },
@@ -753,6 +759,8 @@
753
759
 
754
760
  renderHistory('sttTranscriptionHistory', stt.history?.transcriptions || [], [
755
761
  { label: 'Time', render: i => `<span class="timestamp">${formatRelativeTime(i.timestamp)}</span>` },
762
+ { label: 'Model', render: i => escapeHtml(i.repoId || '-') },
763
+ { label: 'Quant', render: i => escapeHtml(i.quantization || '-') },
756
764
  { label: 'Segments', render: i => i.segmentCount || '-' },
757
765
  { label: 'Text Length', render: i => i.textLength || '-' },
758
766
  { label: 'Duration', render: i => `${(i.durationMs / 1000).toFixed(1)}s` },