@caldrix-ai/caldrix 0.8.0 → 0.9.0

package/CHANGELOG.md

@@ -3,6 +3,12 @@
 Format after [Keep a Changelog](https://keepachangelog.com/); versions are independent semver, tagged
 `caldrix-js-v<version>`.
 
+## 0.9.0 — 2026-11-06
+
+### Changed
+
+- A repo whose `config.json` ties the output head to the embedding (`tie_word_embeddings: true`) but which ships an `lm_head.weight` follows transformers 5's rule at import: a head whose bytes equal the embedding's changes nothing; a head whose bytes differ unties the model, and `Caldrix.load` places and runs the shipped head (before, the embedding was the head and the shipped one was ignored); a head whose shape or type differs from the embedding's is a refusal naming both tensors. The precheck before download fits a tied repo untied when its `model.safetensors.index.json` lists `lm_head.weight` (an equal head is counted one table over), and a load whose import unties the model re-runs the fit before placing it. Pinned by the host crate's native tests (placement of a differing and an equal head, the precheck over a listed head and over the imported tie); not run in a browser: no tied repo shipping a differing head has loaded in one (`inferno/WEB-VERIFICATION.md` records nothing for it).
+
 ## 0.8.0 — 2026-11-06
 
 ### Changed

package/README.md

@@ -8,7 +8,7 @@ This page states only what `inferno/WEB-VERIFICATION.md` records (in
 the Caldrix repository, which is private, so you cannot open it), what the package's own tests run in
 CI, and what the source does.
 
-<!-- host-commit: f39f5b3e7d3c3cf4349d02a1fd4e1042e30efebc -->
+<!-- host-commit: 2608da856bb74ee230b4e294bb36441adc9b5ffc -->
 
 The compiled host in `host/` is built from the `caldrix-web` crate at the commit in the marker above.
 The release workflow overwrites that marker with the commit it builds; the value in a source checkout

package/dist/index.js

@@ -1 +1 @@
-import{a as N,b as i,d as J}from"./chunk-PZGTLL5H.js";function v(e){return typeof e=="object"&&e!==null}function te(e){return v(e)&&typeof e.token=="number"&&typeof e.text=="string"&&typeof e.done=="boolean"&&(e.finishReason===null||e.finishReason==="stop"||e.finishReason==="length"||e.finishReason==="abort")&&typeof e.promptTokens=="number"&&typeof e.completionTokens=="number"&&typeof e.cachedTokens=="number"&&(e.logprobs===void 0||Array.isArray(e.logprobs)&&e.logprobs.every(t=>v(t)&&typeof t.token=="number"&&typeof t.logprob=="number"))}function re(e){return!v(e)||typeof e.url!="string"?!1:e.failed===void 0||v(e.failed)&&typeof e.failed.url=="string"&&typeof e.failed.reason=="string"}function T(e){if(!v(e)||typeof e.id!="number")return!1;switch(e.type){case"ready":return e.id===0&&typeof e.url=="string";case"host":return re(e.hostWasm);case"progress":return typeof e.file=="string"&&typeof e.loaded=="number"&&(e.total===null||typeof e.total=="number")&&typeof e.repo=="string"&&typeof e.revision=="string"&&(e.provenance===void 0||typeof e.provenance=="string");case"loaded":return v(e.doctor)&&v(e.sources)&&(e.fallback===void 0||typeof e.fallback=="string")&&Object.values(e.sources).every(t=>v(t)&&(t.source==="cache"||t.source==="network")&&typeof t.verified=="boolean");case"checked":return e.backend==="webgpu"||e.backend==="wasm";case"fitted":return(e.backend==="webgpu"||e.backend==="wasm")&&(e.maxContext===null||typeof e.maxContext=="number")&&(e.contextClamp===null||typeof e.contextClamp=="string");case"delta":return te(e.delta);case"result":return"value"in e;case"error":return N(e.kind)&&typeof e.message=="string"&&(e.fatal===void 0||typeof e.fatal=="string");default:return!1}}function w(e){let t;try{t=JSON.stringify(e)??String(e)}catch{t=String(e)}return t.length>200&&(t=`${t.slice(0,200)}...`),new i("backend",`the worker sent a malformed message: ${t}`)}var C=(...e)=>new Uint8Array(e),ne={simd:C(0,97,115,109,1,0,0,0,1,5,1,96,0,1,123,3,2,1,0,10,10,1,8,0,65,0,253,15,253,98,11),relaxedSimd:C(0,97,115,109,1,0,0,0,1,5,1,96,0,1,123,3,2,1,0,10,15,1,13,0,65,1,253,15,65,2,253,15,253,128,2,11),bulkMemory:C(0,97,115,109,1,0,0,0,1,4,1,96,0,0,3,2,1,0,5,3,1,0,1,10,14,1,12,0,65,0,65,0,65,0,252,10,0,0,11),threads:C(0,97,115,109,1,0,0,0,1,4,1,96,0,0,3,2,1,0,5,4,1,3,1,1,10,11,1,9,0,65,0,254,16,2,0,26,11),memory64:C(0,97,115,109,1,0,0,0,5,3,1,4,1),tailCall:C(0,97,115,109,1,0,0,0,1,4,1,96,0,0,3,2,1,0,10,6,1,4,0,18,0,11),exceptions:C(0,97,115,109,1,0,0,0,1,4,1,96,0,0,3,2,1,0,10,16,1,14,0,2,105,31,64,1,3,0,0,11,0,11,26,11)};function H(){let e={};for(let[t,o]of Object.entries(ne))e[t]=WebAssembly.validate(o);return e}var oe=1e3,D=Symbol("timed out");async function K(e){let t,o=new Promise(n=>{t=setTimeout(()=>n(D),oe)});try{return await Promise.race([e,o])}finally{clearTimeout(t)}}var X=e=>e instanceof Error?e.message:String(e);function Q(){let e=globalThis.navigator;return typeof e=="object"&&e!==null?e:{}}function P(e){return typeof e=="number"&&Number.isFinite(e)?e:null}function se(){let e={available:!1,maxTextureSize:null,floatColorBuffer:null};try{let t=null;if(typeof OffscreenCanvas<"u"?t=new OffscreenCanvas(1,1):typeof document<"u"&&(t=document.createElement("canvas")),t===null)return e;let o=t,n=o.getContext("webgl2"),r=n??o.getContext("webgl");if(!r)return e;let s=P(r.getParameter(r.MAX_TEXTURE_SIZE)),l=r.getExtension(n?"EXT_color_buffer_float":"WEBGL_color_buffer_float")!==null;return r.getExtension("WEBGL_lose_context")?.loseContext(),{available:!0,maxTextureSize:s,floatColorBuffer:l}}catch{return e}}async function ae(e){let t=e.info??{};if(!e.info&&typeof e.requestAdapterInfo=="function")try{t=await e.requestAdapterInfo()}catch{t={}}let o={},n=[];if(e.limits)for(let r in e.limits){let s=e.limits[r];typeof s=="number"?(o[r]=s,Number.isSafeInteger(s)||n.push(`${r} = ${s} (not a safe integer)`)):typeof s!="function"&&n.push(`${r} = ${JSON.stringify(s)??String(s)} (not a number)`)}return{vendor:t.vendor??"",architecture:t.architecture??"",device:t.device??"",description:t.description??"",features:e.features?[...e.features]:[],limits:o,unrepresentable:n}}async function ie(){let e=Q().gpu;if(!e||typeof e.requestAdapter!="function")return{webgpu:null,available:!1,reason:"no navigator.gpu"};let t;try{let n=await K(e.requestAdapter());if(n===D)return{webgpu:null,available:!1,reason:"requestAdapter timed out"};t=n}catch(n){return{webgpu:null,available:!1,reason:`requestAdapter failed: ${X(n)}`}}if(!t)return{webgpu:null,available:!1,reason:"requestAdapter returned null"};let o=await ae(t);try{let n=await K(t.requestDevice());if(n===D)return{webgpu:o,available:!1,reason:"requestDevice timed out"};n.destroy?.()}catch(n){return{webgpu:o,available:!1,reason:`requestDevice refused: ${X(n)}`}}return{webgpu:o,available:!0}}function le(e){return e.webgpu.available&&e.webgpu.built?"webgpu":"wasm"}function ue(e,t,o){let n=e.webgpu?.unrepresentable??[],r=n.length>0?`; adapter limits ${n.join(", ")}: the host reads them as missing`:"";return de(e,t,o)+r}function de(e,t,o){if(o==="webgpu"){let r=e.webgpu,s=r.features.includes("shader-f16")?" with shader-f16":"";return`WebGPU adapter present (${r.vendor} ${r.architecture})${s}: webgpu`}let n=e.wasm.simd?"with simd128":"without simd128";return e.webgpu===null?`no WebGPU adapter (${t.reason??"unavailable"}): wasm ${n}`:`WebGPU adapter present but unusable (${t.reason}): wasm ${n}`}async function G(){let e=Q(),t=await ie(),o={wasm:H(),threads:{sharedArrayBuffer:typeof SharedArrayBuffer<"u",crossOriginIsolated:globalThis.crossOriginIsolated===!0,hardwareConcurrency:P(e.hardwareConcurrency)},webgl:se(),webgpu:t.webgpu,memory:{deviceMemoryGiB:P(e.deviceMemory)},tiers:{webgpu:t.available?{available:!0,built:!0}:{available:!1,built:!0,reason:t.reason},webgl:{available:!1,built:!1,reason:"not built"},wasm:{available:!0,built:!0}}};o.tiers.webgl.available=o.webgl.available;let n=le(o.tiers);return{...o,chosen:n,why:ue(o,t,n)}}function z(e){let t=e.webgpu===null?null:{vendor:e.webgpu.vendor,architecture:e.webgpu.architecture,device:e.webgpu.device,description:e.webgpu.description,features:e.webgpu.features,limits:Object.fromEntries(Object.entries(e.webgpu.limits).filter(([,o])=>Number.isSafeInteger(o)))};return JSON.stringify({wasm:{simd:e.wasm.simd,relaxedSimd:e.wasm.relaxedSimd,bulkMemory:e.wasm.bulkMemory,threads:e.wasm.threads,memory64:e.wasm.memory64,tailCall:e.wasm.tailCall,exceptions:e.wasm.exceptions},threads:{sharedArrayBuffer:e.threads.sharedArrayBuffer,crossOriginIsolated:e.threads.crossOriginIsolated,hardwareConcurrency:e.threads.hardwareConcurrency},webgpu:t,memory:{deviceMemoryGiB:e.memory.deviceMemoryGiB}})}var ce="https://huggingface.co";function fe(e){return typeof location<"u"?new URL(String(e),location.href):new URL(String(e))}var pe="https://cdn.jsdelivr.net/npm/@caldrix-ai/caldrix@0.8.0/";function S(e){return e instanceof Error?e.message:String(e)}function F(e,t){if(t!==void 0){if(typeof t!="string"&&!(t instanceof URL))throw new i("request",`assets.${e} is a string or a URL, not ${typeof t}`);try{return typeof location<"u"?new URL(t,location.href):new URL(t)}catch{throw new i("request",`assets.${e} ${JSON.stringify(String(t))} is not a URL`)}}}function be(e,t){let o=_(e,t),n=null,r=()=>{n!==null&&URL.revokeObjectURL(n),n=null};try{return!o&&!t?{worker:new Worker(new URL("./worker.js",import.meta.url),{type:"module"}),revoke:r,literal:!0}:o?(n=URL.createObjectURL(new Blob([`import ${JSON.stringify(e.href)};`],{type:"text/javascript"})),{worker:new Worker(n,{type:"module"}),revoke:r}):{worker:new Worker(e,{type:"module"}),revoke:r}}catch(s){throw r(),s}}function me(e){return typeof location<"u"&&e.origin!==location.origin}function _(e,t){return me(e)&&(t||e.protocol==="http:"||e.protocol==="https:")}function q(e){return!e.chosen&&!_(e.url,!1)?`the package's worker (dist/worker.js as the page's bundler emitted it, or ${e.url.href} unbundled)`:`the engine's worker from ${e.url.href}${_(e.url,e.chosen)?" (through a same-origin shim)":""}`}var R=class{constructor(t){this.reason=t}reason};function ge(e,t,o){return new Promise((n,r)=>{let s;try{s=be(e.url,e.chosen)}catch(u){r(new R(`the page could not construct it: ${S(u)}`));return}let{worker:l,revoke:c}=s,b=()=>d(o()),d=(u,a=e.url)=>{if(c(),l.onmessage=null,l.onerror=null,t?.removeEventListener("abort",b),u===null)return n({worker:l,url:a});l.terminate(),r(u)};t?.addEventListener("abort",b,{once:!0}),l.onerror=u=>d(new R(u.message||"it did not load")),l.onmessage=u=>{let a=u.data;if(!T(a)||a.id!==0)return d(w(a));if(a.type==="ready")return s.literal?URL.canParse(a.url)?d(null,new URL(a.url)):d(w(a)):d(null);if(a.type==="error")return d(new R(a.message));d(w(a))}})}async function he(e,t,o,n){let r=[];for(let d of e)try{let{worker:u,url:a}=await ge(d,o,n),f=r[0];return{worker:u,used:d,url:a,...f!==void 0?{failed:{what:q(f.t),reason:f.reason}}:{}}}catch(u){if(!(u instanceof R))throw u;r.push({t:d,reason:u.reason})}let[s,l]=r;if(s!==void 0&&l!==void 0){let d=l.t.url.origin;throw new i("backend",`${q(s.t)} did not start (${s.reason}), nor from the CDN fallback ${l.t.url.href} (${l.reason}); serve dist/ and host/ beside the bundle, set assets.worker to where the page serves them, or allow ${d} in the page's Content-Security-Policy (worker-src blob: for the same-origin shim that starts it, script-src ${d} for the modules it imports, connect-src ${d} for the .wasm)`)}let c=s,b=t&&c.t.source==="bundle"?"; the CDN fallback is off (assets.fallback is false)":"";throw new i("backend",`${q(c.t)} did not start: ${c.reason}${b}; caldrix.js runs inference in a Web Worker and has no main-thread mode`)}function we(e,t,o){let n=o!==void 0?`, and with it the host .wasm from ${o}`:"";return`${e.what} did not start (${e.reason}); it was loaded from the CDN fallback ${t}${n}`}function ke(e){return`the host .wasm could not be fetched from ${e.failed.url} (${e.failed.reason}); it was fetched from the CDN fallback ${e.url}`}function V(e){console.warn(`caldrix.js: ${e}. Serve dist/ and host/ beside the bundle or set assets.worker/assets.hostWasm; assets.fallback: false turns the fallback off.`)}function ye(e){let t={text:e.text,token:e.token,done:e.done};return e.logprobs!==void 0&&(t.logprobs=e.logprobs),e.done&&(e.finishReason!==null&&(t.finishReason=e.finishReason),t.usage={promptTokens:e.promptTokens,completionTokens:e.completionTokens,cachedTokens:e.cachedTokens}),t}var Y=4<<20;async function B(e,t){let o=t.backend??"auto";if(o==="webgl")throw new i("backend","the webgl backend is not built");if(o!=="auto"&&o!=="webgpu"&&o!=="wasm")throw new i("request",`backend ${JSON.stringify(o)} is not auto, webgpu, wasm or webgl`);let n=e.kind==="fit"?"":e.model;if(e.kind!=="fit"&&(typeof n!="string"||n===""))throw new i("request","load takes a model alias or a Hub repo id (owner/name[@revision]), a non-empty string");if(t.file!==void 0&&(typeof t.file!="string"||t.file===""))throw new i("request",`file is the name of a GGUF file in the repo, a non-empty string, not ${JSON.stringify(t.file)?.slice(0,80)}`);let r=F("worker",t.assets?.worker),s=F("hostWasm",t.assets?.hostWasm),l=t.assets?.fallback??!0;if(typeof l!="boolean")throw new i("request",`assets.fallback is true or false, not ${JSON.stringify(l)}`);let c=F("cdn",t.assets?.cdn)??new URL(pe);c.pathname.endsWith("/")||(c.pathname+="/");let b=[r!==void 0?{url:r,chosen:!0,source:"page"}:{url:fe(new URL("./worker.js",import.meta.url)),chosen:!1,source:"bundle"}];l&&r===void 0&&b.push({url:new URL("dist/worker.js",c),chosen:!0,source:"cdn"});let d=l&&s===void 0?new URL("host/caldrix_web_bg.wasm",c).href:void 0,u=t.storage??"auto";if(u!=="auto"&&u!=="f32")throw new i("request",`storage ${JSON.stringify(u)} is not auto or f32`);if(o==="wasm"&&u==="f32")throw new i("request",`storage "f32" is the WebGPU tier's weight storage; backend "wasm" takes none`);if(t.hubBaseUrl!==void 0){let g=t.hubBaseUrl,U=!1;if(typeof g=="string")try{let j=new URL(g);U=j.protocol==="http:"||j.protocol==="https:"}catch{U=!1}if(!U)throw new i("request",`hubBaseUrl must be an absolute http(s) URL, not ${JSON.stringify(g).slice(0,80)}`)}let a=()=>new i("fetch",e.kind==="fit"?"the fit was aborted":`${e.kind==="check"?"checking":"loading"} ${n} was aborted`),f=t.signal;if(f?.aborted)throw a();let h=await G(),m,y;if(o==="auto"&&t.dtype==="bf16"&&h.chosen==="webgpu")m="wasm",y=`bf16 requested: wasm (the WebGPU tier runs the f32 plan only); ${h.why}`;else if(o==="auto")m=h.chosen,y=h.why;else if(o==="webgpu"){if(!h.tiers.webgpu.available)throw new i("backend",`backend "webgpu" was asked for, but the probe found no usable WebGPU adapter: ${h.tiers.webgpu.reason??h.why}`);m="webgpu",y=`backend "webgpu" asked for; ${h.why}`}else m="wasm",y=`backend "wasm" asked for; ${h.why}`;let p=h.webgpu?.unrepresentable??[];if(f?.aborted)throw a();let k=await he(b,!l,f,a),{worker:W}=k,$=k.url,L=o==="auto"&&m==="webgpu"?{auto:!0}:{},O={...s!==void 0?{hostWasm:s.href}:{},...d!==void 0?{cdnWasm:d}:{}},I=z(h),ee=e.kind==="fit"?{id:1,type:"fit",facts:e.facts,backend:m,...L,...t.maxContext!==void 0?{maxContext:t.maxContext}:{},probeJson:I,...O}:{id:1,type:"load",model:e.model,...t.file!==void 0?{file:t.file}:{},probeJson:I,request:{backend:m,...t.dtype!==void 0?{dtype:t.dtype}:{},...t.maxContext!==void 0?{maxContext:t.maxContext}:{},...m==="webgpu"&&u==="f32"?{f16:!1}:{}},...L,...e.kind==="check"?{check:!0}:{},offline:t.offline??!1,hubBaseUrl:t.hubBaseUrl??ce,...O},x={worker:W,used:k.used,workerUrl:$,caps:h,why:y,hostWasm:s,request:ee,signal:f,aborted:a,fellBack:null,wasm:null,noted:g=>x.fellBack===null?g:new i(g.kind,`${g.message} (after a CDN fallback: ${x.fellBack})`,g.missing),host:g=>x.wasm!==null?!1:(x.wasm=g,g.failed!==void 0&&(x.fellBack=ke({...g,failed:g.failed}),V(x.fellBack)),!0),workerError:g=>{let U=g.kind==="probe"&&p.length>0?`; the browser gave ${p.join(", ")}, which the host reads as missing`:"";return new i(g.kind,g.message+U)}};if(k.failed!==void 0&&(x.fellBack=we(k.failed,$.href,s===void 0?d:void 0),V(x.fellBack)),f?.aborted)throw W.terminate(),x.noted(a());return x}function A(e){return e instanceof i?{ok:!1,kind:e.kind,message:e.message}:{ok:!1,kind:"backend",message:S(e)}}function Z(e,t,o){let{worker:n,workerUrl:r,signal:s,request:l}=e;return new Promise(c=>{let b=()=>u(e.aborted()),d=a=>{n.onmessage=null,n.onerror=null,s?.removeEventListener("abort",b),n.terminate(),c(a)},u=a=>d(A(e.noted(a)));if(s?.addEventListener("abort",b,{once:!0}),s?.aborted)return u(e.aborted());n.onerror=a=>u(new i("backend",`the engine's worker from ${r.href} failed: ${a.message||"it did not start"}`)),n.onmessage=a=>{let f=a.data;if(!T(f)||f.id!==l.id)return u(w(f));if(f.type==="host")return e.host(f.hostWasm)?void 0:u(w(f));if(f.type!=="progress")return f.type==="error"?u(e.workerError(f)):f.type!==t?u(w(f)):e.wasm===null?u(w(f)):d(o(f))};try{n.postMessage(l)}catch(a){d(A(a))}})}var M=class{static async load(t,o={}){let n=await B({kind:"load",model:t},o),{worker:r,workerUrl:s,signal:l,request:c,hostWasm:b,caps:d}=n;return new Promise((u,a)=>{let f=()=>m(n.aborted()),h=()=>{r.onmessage=null,r.onerror=null,l?.removeEventListener("abort",f)},m=y=>{h(),r.terminate(),a(n.noted(y))};if(l?.addEventListener("abort",f,{once:!0}),l?.aborted)return m(n.aborted());r.onerror=y=>m(new i("backend",`the engine's worker from ${s.href} failed: ${y.message||"it did not start"}`)),r.onmessage=y=>{let p=y.data;if(!T(p)||p.id!==c.id)return m(w(p));switch(p.type){case"host":return n.host(p.hostWasm)?void 0:m(w(p));case"progress":try{o.onProgress?.({file:p.file,loaded:p.loaded,total:p.total,repo:p.repo,revision:p.revision,...p.provenance!==void 0?{provenance:p.provenance}:{}})}catch(k){m(k instanceof i?k:new i("request",`onProgress threw: ${S(k)}`))}return;case"loaded":{let k=n.wasm;if(k===null)return m(w(p));h();let W=n.used.source,$=b!==void 0||W==="page"&&k.failed===void 0?"page":k.failed!==void 0||W==="cdn"?"cdn":"bundle",L={worker:{source:W,url:s.href},hostWasm:{source:$,url:k.url}},O={...p.doctor,source:p.sources,why:p.fallback!==void 0?`${p.fallback}; ${d.why}`:n.why,assets:L};u(new E(r,O));return}case"error":return m(n.workerError(p));default:return m(w(p))}},r.postMessage(c)})}static async check(t,o={}){let n;try{let r=o??{};n=await B({kind:"check",model:t},{...r.backend!==void 0?{backend:r.backend}:{},...r.dtype!==void 0?{dtype:r.dtype}:{},...r.maxContext!==void 0?{maxContext:r.maxContext}:{},...r.storage!==void 0?{storage:r.storage}:{},...r.hubBaseUrl!==void 0?{hubBaseUrl:r.hubBaseUrl}:{},...r.file!==void 0?{file:r.file}:{},...r.signal!==void 0?{signal:r.signal}:{},...r.assets!==void 0?{assets:r.assets}:{}})}catch(r){return A(r)}return Z(n,"checked",r=>({ok:!0,backend:r.backend}))}static async fit(t,o={}){let n;try{let r=typeof t=="object"&&t!==null?Object.getPrototypeOf(t):void 0;if(r!==Object.prototype&&r!==null)throw new i("request","fit takes a GGUF file's recorded facts, a plain object");let s;try{s=JSON.stringify(t)}catch(b){throw new i("request",`the facts do not serialise: ${S(b)}`)}if(typeof s!="string")throw new i("request","the facts do not serialise to a JSON string");let l=new TextEncoder().encode(s).byteLength;if(l>Y)throw new i("request",`the facts are ${l} bytes once serialised, past the ${Y} a fit takes`);let c=o??{};n=await B({kind:"fit",facts:s},{...c.backend!==void 0?{backend:c.backend}:{},...c.maxContext!==void 0?{maxContext:c.maxContext}:{},...c.signal!==void 0?{signal:c.signal}:{},...c.assets!==void 0?{assets:c.assets}:{}})}catch(r){return A(r)}return Z(n,"fitted",r=>({ok:!0,backend:r.backend,maxContext:r.maxContext,contextClamp:r.contextClamp}))}},E=class{#r;#n;#l=2;#t=!1;#e=null;#i=!1;#o=null;constructor(t,o){this.#r=t,this.#n=o,t.onmessage=n=>this.#d(n.data),t.onerror=n=>this.#a(new i("backend",`the engine's worker failed: ${n.message||"it stopped"}`))}chat(t,o={}){return this.#s(),this.#u({messages:t},o)}complete(t,o={}){this.#s();let n;if(typeof t=="string")n={text:t};else if(typeof t=="object"&&t!==null){let r=t;if(r.text!==void 0&&r.ids!==void 0)throw new i("request","complete takes exactly one of text and ids, not both");if(!Array.isArray(r.ids))throw new i("request","complete takes exactly one of text and ids");n={ids:r.ids}}else throw new i("request","complete takes exactly one of text and ids");return this.#u(n,o)}doctor(){this.#s();let t=Object.fromEntries(Object.entries(this.#n.source).map(([s,l])=>[s,{...l}])),{worker:o,hostWasm:n}=this.#n.assets,r=this.#n.aliases.map(s=>({...s}));return{...this.#n,source:t,aliases:r,assets:{worker:{...o},hostWasm:{...n}}}}dispose(){if(!this.#t){this.#t=!0;try{this.#r.postMessage({id:this.#l++,type:"dispose"})}catch{}this.#r.terminate(),this.#o?.sink(new i("request","the engine was disposed"))}}#s(){if(this.#t)throw new i("request","the engine was disposed; load a new one");if(this.#e)throw this.#e}#a(t){this.#e||this.#t||(this.#e=t,this.#r.terminate(),this.#o?.sink(t))}#d(t){if(this.#t||this.#e)return;if(!T(t))return this.#a(w(t));let o=this.#o;if(o===null||t.id!==o.id)return this.#a(w(t));switch(t.type){case"delta":return o.sink(ye(t.delta));case"error":return t.fatal!==void 0&&(this.#e=J(t.fatal)),o.sink(new i(t.kind,t.message));default:return this.#a(w(t))}}async*#u(t,o){if(this.#s(),this.#i)throw new i("request","a generation is already running on this engine");let{signal:n,...r}=o;if(n?.aborted){yield{text:"",done:!0,finishReason:"abort",usage:{promptTokens:0,completionTokens:0,cachedTokens:0}};return}this.#i=!0;let s=this.#l++,l=[],c=null;this.#o={id:s,sink:a=>{l.push(a),c?.()}};let b=()=>{!this.#t&&!this.#e&&this.#r.postMessage({id:s,type:"abort"})},d=!1,u=async()=>{for(;l.length===0;)await new Promise(a=>c=a);return c=null,l.shift()};try{for(this.#r.postMessage({id:s,type:"generate",prompt:t,options:r}),n?.addEventListener("abort",b,{once:!0});;){let a=await u();if(a instanceof i)throw d=!0,a;if(a.done&&(d=!0),yield a,a.done)return}}finally{if(n?.removeEventListener("abort",b),!d)try{for(b();!(this.#t||this.#e);){let a=await u();if(a instanceof i||a.done)break}}catch{}this.#o=null,this.#i=!1}}};export{M as Caldrix,i as CaldrixError,E as Engine,G as probe};
+import{a as N,b as i,d as J}from"./chunk-PZGTLL5H.js";function v(e){return typeof e=="object"&&e!==null}function te(e){return v(e)&&typeof e.token=="number"&&typeof e.text=="string"&&typeof e.done=="boolean"&&(e.finishReason===null||e.finishReason==="stop"||e.finishReason==="length"||e.finishReason==="abort")&&typeof e.promptTokens=="number"&&typeof e.completionTokens=="number"&&typeof e.cachedTokens=="number"&&(e.logprobs===void 0||Array.isArray(e.logprobs)&&e.logprobs.every(t=>v(t)&&typeof t.token=="number"&&typeof t.logprob=="number"))}function re(e){return!v(e)||typeof e.url!="string"?!1:e.failed===void 0||v(e.failed)&&typeof e.failed.url=="string"&&typeof e.failed.reason=="string"}function T(e){if(!v(e)||typeof e.id!="number")return!1;switch(e.type){case"ready":return e.id===0&&typeof e.url=="string";case"host":return re(e.hostWasm);case"progress":return typeof e.file=="string"&&typeof e.loaded=="number"&&(e.total===null||typeof e.total=="number")&&typeof e.repo=="string"&&typeof e.revision=="string"&&(e.provenance===void 0||typeof e.provenance=="string");case"loaded":return v(e.doctor)&&v(e.sources)&&(e.fallback===void 0||typeof e.fallback=="string")&&Object.values(e.sources).every(t=>v(t)&&(t.source==="cache"||t.source==="network")&&typeof t.verified=="boolean");case"checked":return e.backend==="webgpu"||e.backend==="wasm";case"fitted":return(e.backend==="webgpu"||e.backend==="wasm")&&(e.maxContext===null||typeof e.maxContext=="number")&&(e.contextClamp===null||typeof e.contextClamp=="string");case"delta":return te(e.delta);case"result":return"value"in e;case"error":return N(e.kind)&&typeof e.message=="string"&&(e.fatal===void 0||typeof e.fatal=="string");default:return!1}}function w(e){let t;try{t=JSON.stringify(e)??String(e)}catch{t=String(e)}return t.length>200&&(t=`${t.slice(0,200)}...`),new i("backend",`the worker sent a malformed message: ${t}`)}var C=(...e)=>new Uint8Array(e),ne={simd:C(0,97,115,109,1,0,0,0,1,5,1,96,0,1,123,3,2,1,0,10,10,1,8,0,65,0,253,15,253,98,11),relaxedSimd:C(0,97,115,109,1,0,0,0,1,5,1,96,0,1,123,3,2,1,0,10,15,1,13,0,65,1,253,15,65,2,253,15,253,128,2,11),bulkMemory:C(0,97,115,109,1,0,0,0,1,4,1,96,0,0,3,2,1,0,5,3,1,0,1,10,14,1,12,0,65,0,65,0,65,0,252,10,0,0,11),threads:C(0,97,115,109,1,0,0,0,1,4,1,96,0,0,3,2,1,0,5,4,1,3,1,1,10,11,1,9,0,65,0,254,16,2,0,26,11),memory64:C(0,97,115,109,1,0,0,0,5,3,1,4,1),tailCall:C(0,97,115,109,1,0,0,0,1,4,1,96,0,0,3,2,1,0,10,6,1,4,0,18,0,11),exceptions:C(0,97,115,109,1,0,0,0,1,4,1,96,0,0,3,2,1,0,10,16,1,14,0,2,105,31,64,1,3,0,0,11,0,11,26,11)};function H(){let e={};for(let[t,o]of Object.entries(ne))e[t]=WebAssembly.validate(o);return e}var oe=1e3,D=Symbol("timed out");async function K(e){let t,o=new Promise(n=>{t=setTimeout(()=>n(D),oe)});try{return await Promise.race([e,o])}finally{clearTimeout(t)}}var X=e=>e instanceof Error?e.message:String(e);function Q(){let e=globalThis.navigator;return typeof e=="object"&&e!==null?e:{}}function P(e){return typeof e=="number"&&Number.isFinite(e)?e:null}function se(){let e={available:!1,maxTextureSize:null,floatColorBuffer:null};try{let t=null;if(typeof OffscreenCanvas<"u"?t=new OffscreenCanvas(1,1):typeof document<"u"&&(t=document.createElement("canvas")),t===null)return e;let o=t,n=o.getContext("webgl2"),r=n??o.getContext("webgl");if(!r)return e;let s=P(r.getParameter(r.MAX_TEXTURE_SIZE)),l=r.getExtension(n?"EXT_color_buffer_float":"WEBGL_color_buffer_float")!==null;return r.getExtension("WEBGL_lose_context")?.loseContext(),{available:!0,maxTextureSize:s,floatColorBuffer:l}}catch{return e}}async function ae(e){let t=e.info??{};if(!e.info&&typeof e.requestAdapterInfo=="function")try{t=await e.requestAdapterInfo()}catch{t={}}let o={},n=[];if(e.limits)for(let r in e.limits){let s=e.limits[r];typeof s=="number"?(o[r]=s,Number.isSafeInteger(s)||n.push(`${r} = ${s} (not a safe integer)`)):typeof s!="function"&&n.push(`${r} = ${JSON.stringify(s)??String(s)} (not a number)`)}return{vendor:t.vendor??"",architecture:t.architecture??"",device:t.device??"",description:t.description??"",features:e.features?[...e.features]:[],limits:o,unrepresentable:n}}async function ie(){let e=Q().gpu;if(!e||typeof e.requestAdapter!="function")return{webgpu:null,available:!1,reason:"no navigator.gpu"};let t;try{let n=await K(e.requestAdapter());if(n===D)return{webgpu:null,available:!1,reason:"requestAdapter timed out"};t=n}catch(n){return{webgpu:null,available:!1,reason:`requestAdapter failed: ${X(n)}`}}if(!t)return{webgpu:null,available:!1,reason:"requestAdapter returned null"};let o=await ae(t);try{let n=await K(t.requestDevice());if(n===D)return{webgpu:o,available:!1,reason:"requestDevice timed out"};n.destroy?.()}catch(n){return{webgpu:o,available:!1,reason:`requestDevice refused: ${X(n)}`}}return{webgpu:o,available:!0}}function le(e){return e.webgpu.available&&e.webgpu.built?"webgpu":"wasm"}function ue(e,t,o){let n=e.webgpu?.unrepresentable??[],r=n.length>0?`; adapter limits ${n.join(", ")}: the host reads them as missing`:"";return de(e,t,o)+r}function de(e,t,o){if(o==="webgpu"){let r=e.webgpu,s=r.features.includes("shader-f16")?" with shader-f16":"";return`WebGPU adapter present (${r.vendor} ${r.architecture})${s}: webgpu`}let n=e.wasm.simd?"with simd128":"without simd128";return e.webgpu===null?`no WebGPU adapter (${t.reason??"unavailable"}): wasm ${n}`:`WebGPU adapter present but unusable (${t.reason}): wasm ${n}`}async function G(){let e=Q(),t=await ie(),o={wasm:H(),threads:{sharedArrayBuffer:typeof SharedArrayBuffer<"u",crossOriginIsolated:globalThis.crossOriginIsolated===!0,hardwareConcurrency:P(e.hardwareConcurrency)},webgl:se(),webgpu:t.webgpu,memory:{deviceMemoryGiB:P(e.deviceMemory)},tiers:{webgpu:t.available?{available:!0,built:!0}:{available:!1,built:!0,reason:t.reason},webgl:{available:!1,built:!1,reason:"not built"},wasm:{available:!0,built:!0}}};o.tiers.webgl.available=o.webgl.available;let n=le(o.tiers);return{...o,chosen:n,why:ue(o,t,n)}}function z(e){let t=e.webgpu===null?null:{vendor:e.webgpu.vendor,architecture:e.webgpu.architecture,device:e.webgpu.device,description:e.webgpu.description,features:e.webgpu.features,limits:Object.fromEntries(Object.entries(e.webgpu.limits).filter(([,o])=>Number.isSafeInteger(o)))};return JSON.stringify({wasm:{simd:e.wasm.simd,relaxedSimd:e.wasm.relaxedSimd,bulkMemory:e.wasm.bulkMemory,threads:e.wasm.threads,memory64:e.wasm.memory64,tailCall:e.wasm.tailCall,exceptions:e.wasm.exceptions},threads:{sharedArrayBuffer:e.threads.sharedArrayBuffer,crossOriginIsolated:e.threads.crossOriginIsolated,hardwareConcurrency:e.threads.hardwareConcurrency},webgpu:t,memory:{deviceMemoryGiB:e.memory.deviceMemoryGiB}})}var ce="https://huggingface.co";function fe(e){return typeof location<"u"?new URL(String(e),location.href):new URL(String(e))}var pe="https://cdn.jsdelivr.net/npm/@caldrix-ai/caldrix@0.9.0/";function S(e){return e instanceof Error?e.message:String(e)}function F(e,t){if(t!==void 0){if(typeof t!="string"&&!(t instanceof URL))throw new i("request",`assets.${e} is a string or a URL, not ${typeof t}`);try{return typeof location<"u"?new URL(t,location.href):new URL(t)}catch{throw new i("request",`assets.${e} ${JSON.stringify(String(t))} is not a URL`)}}}function be(e,t){let o=_(e,t),n=null,r=()=>{n!==null&&URL.revokeObjectURL(n),n=null};try{return!o&&!t?{worker:new Worker(new URL("./worker.js",import.meta.url),{type:"module"}),revoke:r,literal:!0}:o?(n=URL.createObjectURL(new Blob([`import ${JSON.stringify(e.href)};`],{type:"text/javascript"})),{worker:new Worker(n,{type:"module"}),revoke:r}):{worker:new Worker(e,{type:"module"}),revoke:r}}catch(s){throw r(),s}}function me(e){return typeof location<"u"&&e.origin!==location.origin}function _(e,t){return me(e)&&(t||e.protocol==="http:"||e.protocol==="https:")}function q(e){return!e.chosen&&!_(e.url,!1)?`the package's worker (dist/worker.js as the page's bundler emitted it, or ${e.url.href} unbundled)`:`the engine's worker from ${e.url.href}${_(e.url,e.chosen)?" (through a same-origin shim)":""}`}var R=class{constructor(t){this.reason=t}reason};function ge(e,t,o){return new Promise((n,r)=>{let s;try{s=be(e.url,e.chosen)}catch(u){r(new R(`the page could not construct it: ${S(u)}`));return}let{worker:l,revoke:c}=s,b=()=>d(o()),d=(u,a=e.url)=>{if(c(),l.onmessage=null,l.onerror=null,t?.removeEventListener("abort",b),u===null)return n({worker:l,url:a});l.terminate(),r(u)};t?.addEventListener("abort",b,{once:!0}),l.onerror=u=>d(new R(u.message||"it did not load")),l.onmessage=u=>{let a=u.data;if(!T(a)||a.id!==0)return d(w(a));if(a.type==="ready")return s.literal?URL.canParse(a.url)?d(null,new URL(a.url)):d(w(a)):d(null);if(a.type==="error")return d(new R(a.message));d(w(a))}})}async function he(e,t,o,n){let r=[];for(let d of e)try{let{worker:u,url:a}=await ge(d,o,n),f=r[0];return{worker:u,used:d,url:a,...f!==void 0?{failed:{what:q(f.t),reason:f.reason}}:{}}}catch(u){if(!(u instanceof R))throw u;r.push({t:d,reason:u.reason})}let[s,l]=r;if(s!==void 0&&l!==void 0){let d=l.t.url.origin;throw new i("backend",`${q(s.t)} did not start (${s.reason}), nor from the CDN fallback ${l.t.url.href} (${l.reason}); serve dist/ and host/ beside the bundle, set assets.worker to where the page serves them, or allow ${d} in the page's Content-Security-Policy (worker-src blob: for the same-origin shim that starts it, script-src ${d} for the modules it imports, connect-src ${d} for the .wasm)`)}let c=s,b=t&&c.t.source==="bundle"?"; the CDN fallback is off (assets.fallback is false)":"";throw new i("backend",`${q(c.t)} did not start: ${c.reason}${b}; caldrix.js runs inference in a Web Worker and has no main-thread mode`)}function we(e,t,o){let n=o!==void 0?`, and with it the host .wasm from ${o}`:"";return`${e.what} did not start (${e.reason}); it was loaded from the CDN fallback ${t}${n}`}function ke(e){return`the host .wasm could not be fetched from ${e.failed.url} (${e.failed.reason}); it was fetched from the CDN fallback ${e.url}`}function V(e){console.warn(`caldrix.js: ${e}. Serve dist/ and host/ beside the bundle or set assets.worker/assets.hostWasm; assets.fallback: false turns the fallback off.`)}function ye(e){let t={text:e.text,token:e.token,done:e.done};return e.logprobs!==void 0&&(t.logprobs=e.logprobs),e.done&&(e.finishReason!==null&&(t.finishReason=e.finishReason),t.usage={promptTokens:e.promptTokens,completionTokens:e.completionTokens,cachedTokens:e.cachedTokens}),t}var Y=4<<20;async function B(e,t){let o=t.backend??"auto";if(o==="webgl")throw new i("backend","the webgl backend is not built");if(o!=="auto"&&o!=="webgpu"&&o!=="wasm")throw new i("request",`backend ${JSON.stringify(o)} is not auto, webgpu, wasm or webgl`);let n=e.kind==="fit"?"":e.model;if(e.kind!=="fit"&&(typeof n!="string"||n===""))throw new i("request","load takes a model alias or a Hub repo id (owner/name[@revision]), a non-empty string");if(t.file!==void 0&&(typeof t.file!="string"||t.file===""))throw new i("request",`file is the name of a GGUF file in the repo, a non-empty string, not ${JSON.stringify(t.file)?.slice(0,80)}`);let r=F("worker",t.assets?.worker),s=F("hostWasm",t.assets?.hostWasm),l=t.assets?.fallback??!0;if(typeof l!="boolean")throw new i("request",`assets.fallback is true or false, not ${JSON.stringify(l)}`);let c=F("cdn",t.assets?.cdn)??new URL(pe);c.pathname.endsWith("/")||(c.pathname+="/");let b=[r!==void 0?{url:r,chosen:!0,source:"page"}:{url:fe(new URL("./worker.js",import.meta.url)),chosen:!1,source:"bundle"}];l&&r===void 0&&b.push({url:new URL("dist/worker.js",c),chosen:!0,source:"cdn"});let d=l&&s===void 0?new URL("host/caldrix_web_bg.wasm",c).href:void 0,u=t.storage??"auto";if(u!=="auto"&&u!=="f32")throw new i("request",`storage ${JSON.stringify(u)} is not auto or f32`);if(o==="wasm"&&u==="f32")throw new i("request",`storage "f32" is the WebGPU tier's weight storage; backend "wasm" takes none`);if(t.hubBaseUrl!==void 0){let g=t.hubBaseUrl,U=!1;if(typeof g=="string")try{let j=new URL(g);U=j.protocol==="http:"||j.protocol==="https:"}catch{U=!1}if(!U)throw new i("request",`hubBaseUrl must be an absolute http(s) URL, not ${JSON.stringify(g).slice(0,80)}`)}let a=()=>new i("fetch",e.kind==="fit"?"the fit was aborted":`${e.kind==="check"?"checking":"loading"} ${n} was aborted`),f=t.signal;if(f?.aborted)throw a();let h=await G(),m,y;if(o==="auto"&&t.dtype==="bf16"&&h.chosen==="webgpu")m="wasm",y=`bf16 requested: wasm (the WebGPU tier runs the f32 plan only); ${h.why}`;else if(o==="auto")m=h.chosen,y=h.why;else if(o==="webgpu"){if(!h.tiers.webgpu.available)throw new i("backend",`backend "webgpu" was asked for, but the probe found no usable WebGPU adapter: ${h.tiers.webgpu.reason??h.why}`);m="webgpu",y=`backend "webgpu" asked for; ${h.why}`}else m="wasm",y=`backend "wasm" asked for; ${h.why}`;let p=h.webgpu?.unrepresentable??[];if(f?.aborted)throw a();let k=await he(b,!l,f,a),{worker:W}=k,$=k.url,L=o==="auto"&&m==="webgpu"?{auto:!0}:{},O={...s!==void 0?{hostWasm:s.href}:{},...d!==void 0?{cdnWasm:d}:{}},I=z(h),ee=e.kind==="fit"?{id:1,type:"fit",facts:e.facts,backend:m,...L,...t.maxContext!==void 0?{maxContext:t.maxContext}:{},probeJson:I,...O}:{id:1,type:"load",model:e.model,...t.file!==void 0?{file:t.file}:{},probeJson:I,request:{backend:m,...t.dtype!==void 0?{dtype:t.dtype}:{},...t.maxContext!==void 0?{maxContext:t.maxContext}:{},...m==="webgpu"&&u==="f32"?{f16:!1}:{}},...L,...e.kind==="check"?{check:!0}:{},offline:t.offline??!1,hubBaseUrl:t.hubBaseUrl??ce,...O},x={worker:W,used:k.used,workerUrl:$,caps:h,why:y,hostWasm:s,request:ee,signal:f,aborted:a,fellBack:null,wasm:null,noted:g=>x.fellBack===null?g:new i(g.kind,`${g.message} (after a CDN fallback: ${x.fellBack})`,g.missing),host:g=>x.wasm!==null?!1:(x.wasm=g,g.failed!==void 0&&(x.fellBack=ke({...g,failed:g.failed}),V(x.fellBack)),!0),workerError:g=>{let U=g.kind==="probe"&&p.length>0?`; the browser gave ${p.join(", ")}, which the host reads as missing`:"";return new i(g.kind,g.message+U)}};if(k.failed!==void 0&&(x.fellBack=we(k.failed,$.href,s===void 0?d:void 0),V(x.fellBack)),f?.aborted)throw W.terminate(),x.noted(a());return x}function A(e){return e instanceof i?{ok:!1,kind:e.kind,message:e.message}:{ok:!1,kind:"backend",message:S(e)}}function Z(e,t,o){let{worker:n,workerUrl:r,signal:s,request:l}=e;return new Promise(c=>{let b=()=>u(e.aborted()),d=a=>{n.onmessage=null,n.onerror=null,s?.removeEventListener("abort",b),n.terminate(),c(a)},u=a=>d(A(e.noted(a)));if(s?.addEventListener("abort",b,{once:!0}),s?.aborted)return u(e.aborted());n.onerror=a=>u(new i("backend",`the engine's worker from ${r.href} failed: ${a.message||"it did not start"}`)),n.onmessage=a=>{let f=a.data;if(!T(f)||f.id!==l.id)return u(w(f));if(f.type==="host")return e.host(f.hostWasm)?void 0:u(w(f));if(f.type!=="progress")return f.type==="error"?u(e.workerError(f)):f.type!==t?u(w(f)):e.wasm===null?u(w(f)):d(o(f))};try{n.postMessage(l)}catch(a){d(A(a))}})}var M=class{static async load(t,o={}){let n=await B({kind:"load",model:t},o),{worker:r,workerUrl:s,signal:l,request:c,hostWasm:b,caps:d}=n;return new Promise((u,a)=>{let f=()=>m(n.aborted()),h=()=>{r.onmessage=null,r.onerror=null,l?.removeEventListener("abort",f)},m=y=>{h(),r.terminate(),a(n.noted(y))};if(l?.addEventListener("abort",f,{once:!0}),l?.aborted)return m(n.aborted());r.onerror=y=>m(new i("backend",`the engine's worker from ${s.href} failed: ${y.message||"it did not start"}`)),r.onmessage=y=>{let p=y.data;if(!T(p)||p.id!==c.id)return m(w(p));switch(p.type){case"host":return n.host(p.hostWasm)?void 0:m(w(p));case"progress":try{o.onProgress?.({file:p.file,loaded:p.loaded,total:p.total,repo:p.repo,revision:p.revision,...p.provenance!==void 0?{provenance:p.provenance}:{}})}catch(k){m(k instanceof i?k:new i("request",`onProgress threw: ${S(k)}`))}return;case"loaded":{let k=n.wasm;if(k===null)return m(w(p));h();let W=n.used.source,$=b!==void 0||W==="page"&&k.failed===void 0?"page":k.failed!==void 0||W==="cdn"?"cdn":"bundle",L={worker:{source:W,url:s.href},hostWasm:{source:$,url:k.url}},O={...p.doctor,source:p.sources,why:p.fallback!==void 0?`${p.fallback}; ${d.why}`:n.why,assets:L};u(new E(r,O));return}case"error":return m(n.workerError(p));default:return m(w(p))}},r.postMessage(c)})}static async check(t,o={}){let n;try{let r=o??{};n=await B({kind:"check",model:t},{...r.backend!==void 0?{backend:r.backend}:{},...r.dtype!==void 0?{dtype:r.dtype}:{},...r.maxContext!==void 0?{maxContext:r.maxContext}:{},...r.storage!==void 0?{storage:r.storage}:{},...r.hubBaseUrl!==void 0?{hubBaseUrl:r.hubBaseUrl}:{},...r.file!==void 0?{file:r.file}:{},...r.signal!==void 0?{signal:r.signal}:{},...r.assets!==void 0?{assets:r.assets}:{}})}catch(r){return A(r)}return Z(n,"checked",r=>({ok:!0,backend:r.backend}))}static async fit(t,o={}){let n;try{let r=typeof t=="object"&&t!==null?Object.getPrototypeOf(t):void 0;if(r!==Object.prototype&&r!==null)throw new i("request","fit takes a GGUF file's recorded facts, a plain object");let s;try{s=JSON.stringify(t)}catch(b){throw new i("request",`the facts do not serialise: ${S(b)}`)}if(typeof s!="string")throw new i("request","the facts do not serialise to a JSON string");let l=new TextEncoder().encode(s).byteLength;if(l>Y)throw new i("request",`the facts are ${l} bytes once serialised, past the ${Y} a fit takes`);let c=o??{};n=await B({kind:"fit",facts:s},{...c.backend!==void 0?{backend:c.backend}:{},...c.maxContext!==void 0?{maxContext:c.maxContext}:{},...c.signal!==void 0?{signal:c.signal}:{},...c.assets!==void 0?{assets:c.assets}:{}})}catch(r){return A(r)}return Z(n,"fitted",r=>({ok:!0,backend:r.backend,maxContext:r.maxContext,contextClamp:r.contextClamp}))}},E=class{#r;#n;#l=2;#t=!1;#e=null;#i=!1;#o=null;constructor(t,o){this.#r=t,this.#n=o,t.onmessage=n=>this.#d(n.data),t.onerror=n=>this.#a(new i("backend",`the engine's worker failed: ${n.message||"it stopped"}`))}chat(t,o={}){return this.#s(),this.#u({messages:t},o)}complete(t,o={}){this.#s();let n;if(typeof t=="string")n={text:t};else if(typeof t=="object"&&t!==null){let r=t;if(r.text!==void 0&&r.ids!==void 0)throw new i("request","complete takes exactly one of text and ids, not both");if(!Array.isArray(r.ids))throw new i("request","complete takes exactly one of text and ids");n={ids:r.ids}}else throw new i("request","complete takes exactly one of text and ids");return this.#u(n,o)}doctor(){this.#s();let t=Object.fromEntries(Object.entries(this.#n.source).map(([s,l])=>[s,{...l}])),{worker:o,hostWasm:n}=this.#n.assets,r=this.#n.aliases.map(s=>({...s}));return{...this.#n,source:t,aliases:r,assets:{worker:{...o},hostWasm:{...n}}}}dispose(){if(!this.#t){this.#t=!0;try{this.#r.postMessage({id:this.#l++,type:"dispose"})}catch{}this.#r.terminate(),this.#o?.sink(new i("request","the engine was disposed"))}}#s(){if(this.#t)throw new i("request","the engine was disposed; load a new one");if(this.#e)throw this.#e}#a(t){this.#e||this.#t||(this.#e=t,this.#r.terminate(),this.#o?.sink(t))}#d(t){if(this.#t||this.#e)return;if(!T(t))return this.#a(w(t));let o=this.#o;if(o===null||t.id!==o.id)return this.#a(w(t));switch(t.type){case"delta":return o.sink(ye(t.delta));case"error":return t.fatal!==void 0&&(this.#e=J(t.fatal)),o.sink(new i(t.kind,t.message));default:return this.#a(w(t))}}async*#u(t,o){if(this.#s(),this.#i)throw new i("request","a generation is already running on this engine");let{signal:n,...r}=o;if(n?.aborted){yield{text:"",done:!0,finishReason:"abort",usage:{promptTokens:0,completionTokens:0,cachedTokens:0}};return}this.#i=!0;let s=this.#l++,l=[],c=null;this.#o={id:s,sink:a=>{l.push(a),c?.()}};let b=()=>{!this.#t&&!this.#e&&this.#r.postMessage({id:s,type:"abort"})},d=!1,u=async()=>{for(;l.length===0;)await new Promise(a=>c=a);return c=null,l.shift()};try{for(this.#r.postMessage({id:s,type:"generate",prompt:t,options:r}),n?.addEventListener("abort",b,{once:!0});;){let a=await u();if(a instanceof i)throw d=!0,a;if(a.done&&(d=!0),yield a,a.done)return}}finally{if(n?.removeEventListener("abort",b),!d)try{for(b();!(this.#t||this.#e);){let a=await u();if(a instanceof i||a.done)break}}catch{}this.#o=null,this.#i=!1}}};export{M as Caldrix,i as CaldrixError,E as Engine,G as probe};

package/host/caldrix_web.d.ts

@@ -119,11 +119,14 @@ export class CaldrixWeb {
     loadWebGpu(config_json: Uint8Array, tokenizer_file: string, tokenizer: Uint8Array, merges: Uint8Array | null | undefined, added_tokens: Uint8Array | null | undefined, tokenizer_config: Uint8Array | null | undefined, chat_template: Uint8Array | null | undefined, files: Array<any>, request_json: string, adapter_config?: Uint8Array | null, adapter_files?: Array<any> | null): any;
     /**
      * Prechecks again on the real bytes, imports, lowers and places into the host's heap; returns
-     * the module bytes for the page to instantiate over [`memory`]. With `adapter_config` and
-     * `adapter_files` (a LoRA adapter repo), or a config carrying a `quantization_config` (a
-     * bitsandbytes nf4 repo), the model is imported densified at the request's plan dtype, then
-     * prechecked over the imported dense descriptor and the files' real lengths, lowered and
-     * placed tensor by tensor; the base's files and the adapter's are freed before this returns.
+     * the module bytes for the page to instantiate over [`memory`]. When the import decides another
+     * tie than the config states (a tied config shipping a differing `lm_head.weight`), the fit is
+     * re-run over the imported descriptor ([`precheck_imported`]) before placement. With
+     * `adapter_config` and `adapter_files` (a LoRA adapter repo), or a config carrying a
+     * `quantization_config` (a bitsandbytes nf4 repo), the model is imported densified at the
+     * request's plan dtype, then prechecked over the imported dense descriptor and the files' real
+     * lengths, lowered and placed tensor by tensor; the base's files and the adapter's are freed
+     * before this returns.
      * `tokenizer_file` names what `tokenizer` holds: `tokenizer.json`, `tokenizer.model`, or
      * `vocab.json` (with the repo's `merges.txt` as `merges`, required with it and refused with
      * anything else); a `tokenizer.model` or `vocab.json` repo's `added_tokens.json` is

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@caldrix-ai/caldrix",
-  "version": "0.8.0",
+  "version": "0.9.0",
   "private": false,
   "description": "Run a language model in the browser, in a worker, on WebAssembly or WebGPU; see the README for what has run.",
   "license": "Apache-2.0",