@qvac/harness 0.5.0 → 0.6.0

package/dist-lib/lib/serve.js

@@ -191,6 +191,18 @@ export function serveHarness(stream, harness) {
             await harness.warmModel({ model, agent: agent ?? {} });
             return { ok: true };
         },
+        loadModel: async ({ model, agent }) => {
+            if (!harness.loadModel) {
+                throw new Error('harness does not support loadModel');
+            }
+            return harness.loadModel({ model, agent: agent ?? {} });
+        },
+        unloadModel: async ({ model }) => {
+            if (!harness.unloadModel) {
+                throw new Error('harness does not support unloadModel');
+            }
+            return harness.unloadModel({ model });
+        },
         modelEvents: () => modelEvents(harness),
         completion: completion(harness)
     });

package/dist-lib/schema/harness.js

@@ -721,6 +721,21 @@ export function registerHarnessTypes(ns) {
         name: 'set-model-idle-timeout-request',
         fields: [{ name: 'ms', type: 'uint', required: true }]
     });
+    ns.register({
+        name: 'load-model-response',
+        fields: [
+            { name: 'loaded', type: 'bool', required: true },
+            { name: 'reason', type: 'string' }
+        ]
+    });
+    ns.register({
+        name: 'unload-model-request',
+        fields: [{ name: 'model', type: 'string', required: true }]
+    });
+    ns.register({
+        name: 'unload-model-response',
+        fields: [{ name: 'unloaded', type: 'bool', required: true }]
+    });
 }
 export function registerHarnessApi(api) {
     api.register({
@@ -878,4 +893,14 @@ export function registerHarnessApi(api) {
         request: { name: '@harness/set-model-idle-timeout-request', stream: false },
         response: { name: '@harness/ok-response', stream: false }
     });
+    api.register({
+        name: 'load-model',
+        request: { name: '@harness/warm-model-request', stream: false },
+        response: { name: '@harness/load-model-response', stream: false }
+    });
+    api.register({
+        name: 'unload-model',
+        request: { name: '@harness/unload-model-request', stream: false },
+        response: { name: '@harness/unload-model-response', stream: false }
+    });
 }

package/dist-lib/spec/bind.js

@@ -133,4 +133,10 @@ export function bindApi(rpc, api) {
   if (api.setModelIdleTimeout) {
     rpc.onSetModelIdleTimeout((req) => api.setModelIdleTimeout(req));
   }
+  if (api.loadModel) {
+    rpc.onLoadModel((req) => api.loadModel(req));
+  }
+  if (api.unloadModel) {
+    rpc.onUnloadModel((req) => api.unloadModel(req));
+  }
 }

package/dist-lib/spec/calls.js

@@ -41,5 +41,7 @@ export function harnessClient(rpc) {
     detectLanguage: (input) => rpc.detectLanguage(input),
     previewModelSource: (input) => rpc.previewModelSource(input),
     setModelIdleTimeout: (input) => rpc.setModelIdleTimeout(input),
+    loadModel: (input) => rpc.loadModel(input),
+    unloadModel: (input) => rpc.unloadModel(input),
   };
 }

package/dist-lib/spec/hrpc/hrpc.json

@@ -403,6 +403,32 @@
         "stream": false
       },
       "version": 1
+    },
+    {
+      "id": 31,
+      "name": "@harness/load-model",
+      "request": {
+        "name": "@harness/warm-model-request",
+        "stream": false
+      },
+      "response": {
+        "name": "@harness/load-model-response",
+        "stream": false
+      },
+      "version": 1
+    },
+    {
+      "id": 32,
+      "name": "@harness/unload-model",
+      "request": {
+        "name": "@harness/unload-model-request",
+        "stream": false
+      },
+      "response": {
+        "name": "@harness/unload-model-response",
+        "stream": false
+      },
+      "version": 1
     }
   ]
 }

package/dist-lib/spec/hrpc/index.js

@@ -67,7 +67,11 @@ const methods = new Map([
   ['@harness/preview-model-source', 29],
   [29, '@harness/preview-model-source'],
   ['@harness/set-model-idle-timeout', 30],
-  [30, '@harness/set-model-idle-timeout']
+  [30, '@harness/set-model-idle-timeout'],
+  ['@harness/load-model', 31],
+  [31, '@harness/load-model'],
+  ['@harness/unload-model', 32],
+  [32, '@harness/unload-model']
 ])
 
 class HRPC {
@@ -105,7 +109,9 @@ class HRPC {
       ['@harness/delete-kv-cache', getEncoding('@harness/kv-cache-delete-request')],
       ['@harness/detect-language', getEncoding('@harness/detect-language-request')],
       ['@harness/preview-model-source', getEncoding('@harness/preview-model-source-request')],
-      ['@harness/set-model-idle-timeout', getEncoding('@harness/set-model-idle-timeout-request')]
+      ['@harness/set-model-idle-timeout', getEncoding('@harness/set-model-idle-timeout-request')],
+      ['@harness/load-model', getEncoding('@harness/warm-model-request')],
+      ['@harness/unload-model', getEncoding('@harness/unload-model-request')]
     ])
     this._responseEncodings = new Map([
       ['@harness/completion', getEncoding('@harness/completion-event')],
@@ -138,7 +144,9 @@ class HRPC {
       ['@harness/delete-kv-cache', getEncoding('@harness/kv-cache-deletion-response')],
       ['@harness/detect-language', getEncoding('@harness/detect-language-response')],
       ['@harness/preview-model-source', getEncoding('@harness/preview-model-source-response')],
-      ['@harness/set-model-idle-timeout', getEncoding('@harness/ok-response')]
+      ['@harness/set-model-idle-timeout', getEncoding('@harness/ok-response')],
+      ['@harness/load-model', getEncoding('@harness/load-model-response')],
+      ['@harness/unload-model', getEncoding('@harness/unload-model-response')]
     ])
     this._rpc = new RPC(stream, async (req) => {
       const command = methods.get(req.command)
@@ -360,6 +368,14 @@ class HRPC {
     return this._call('@harness/set-model-idle-timeout', args)
   }
 
+  async loadModel(args) {
+    return this._call('@harness/load-model', args)
+  }
+
+  async unloadModel(args) {
+    return this._call('@harness/unload-model', args)
+  }
+
   onCompletion(responseFn) {
     this._handlers['@harness/completion'] = responseFn
   }
@@ -484,6 +500,14 @@ class HRPC {
     this._handlers['@harness/set-model-idle-timeout'] = responseFn
   }
 
+  onLoadModel(responseFn) {
+    this._handlers['@harness/load-model'] = responseFn
+  }
+
+  onUnloadModel(responseFn) {
+    this._handlers['@harness/unload-model'] = responseFn
+  }
+
   _requestIsStream(command) {
     return [
       '@harness/completion',

package/dist-lib/spec/hrpc/messages.js

@@ -1,12 +1,12 @@
 // This file is autogenerated by the hyperschema compiler
-// Schema Version: 15
+// Schema Version: 16
 /* eslint-disable camelcase */
 /* eslint-disable quotes */
 /* eslint-disable space-before-function-paren */
 
 import { c } from 'hyperschema/runtime'
 
-const VERSION = 15
+const VERSION = 16
 
 // eslint-disable-next-line no-unused-vars
 let version = VERSION
@@ -3365,6 +3365,66 @@ const encoding88 = {
   }
 }
 
+// @harness/load-model-response
+const encoding89 = {
+  preencode(state, m) {
+    state.end++ // max flag is 2 so always one byte
+
+    if (version >= 16 && m.reason) c.string.preencode(state, m.reason)
+  },
+  encode(state, m) {
+    const flags = ((version >= 16 && m.loaded) ? 1 : 0) | ((version >= 16 && m.reason) ? 2 : 0)
+
+    c.uint.encode(state, flags)
+
+    if (version >= 16 && m.reason) c.string.encode(state, m.reason)
+  },
+  decode(state) {
+    const flags = c.uint.decode(state)
+
+    return {
+      loaded: (version >= 16 && (flags & 1) !== 0),
+      reason: (version >= 16 && (flags & 2) !== 0) ? c.string.decode(state) : null
+    }
+  }
+}
+
+// @harness/unload-model-request
+const encoding90 = {
+  preencode(state, m) {
+    c.string.preencode(state, m.model)
+  },
+  encode(state, m) {
+    c.string.encode(state, m.model)
+  },
+  decode(state) {
+    const r0 = c.string.decode(state)
+
+    return {
+      model: r0
+    }
+  }
+}
+
+// @harness/unload-model-response
+const encoding91 = {
+  preencode(state, m) {
+    state.end++ // max flag is 1 so always one byte
+  },
+  encode(state, m) {
+    const flags = (version >= 16 && m.unloaded) ? 1 : 0
+
+    c.uint.encode(state, flags)
+  },
+  decode(state) {
+    const flags = c.uint.decode(state)
+
+    return {
+      unloaded: (version >= 16 && (flags & 1) !== 0)
+    }
+  }
+}
+
 // @harness/completion-frame.grantedDrives, deferred due to recusive use
 const encoding11_23 = c.array(c.frame(encoding60))
 // @harness/skill.setupV2, deferred due to recusive use
@@ -3605,6 +3665,12 @@ function getEncoding(name) {
       return encoding87
     case '@harness/set-model-idle-timeout-request':
       return encoding88
+    case '@harness/load-model-response':
+      return encoding89
+    case '@harness/unload-model-request':
+      return encoding90
+    case '@harness/unload-model-response':
+      return encoding91
     default:
       throw new Error('Encoder not found ' + name)
   }

package/dist-lib/spec/hyperschema/index.js

@@ -1,12 +1,12 @@
 // This file is autogenerated by the hyperschema compiler
-// Schema Version: 15
+// Schema Version: 16
 /* eslint-disable camelcase */
 /* eslint-disable quotes */
 /* eslint-disable space-before-function-paren */
 
 import { c } from 'hyperschema/runtime'
 
-const VERSION = 15
+const VERSION = 16
 
 // eslint-disable-next-line no-unused-vars
 let version = VERSION
@@ -3365,6 +3365,66 @@ const encoding88 = {
   }
 }
 
+// @harness/load-model-response
+const encoding89 = {
+  preencode(state, m) {
+    state.end++ // max flag is 2 so always one byte
+
+    if (version >= 16 && m.reason) c.string.preencode(state, m.reason)
+  },
+  encode(state, m) {
+    const flags = ((version >= 16 && m.loaded) ? 1 : 0) | ((version >= 16 && m.reason) ? 2 : 0)
+
+    c.uint.encode(state, flags)
+
+    if (version >= 16 && m.reason) c.string.encode(state, m.reason)
+  },
+  decode(state) {
+    const flags = c.uint.decode(state)
+
+    return {
+      loaded: (version >= 16 && (flags & 1) !== 0),
+      reason: (version >= 16 && (flags & 2) !== 0) ? c.string.decode(state) : null
+    }
+  }
+}
+
+// @harness/unload-model-request
+const encoding90 = {
+  preencode(state, m) {
+    c.string.preencode(state, m.model)
+  },
+  encode(state, m) {
+    c.string.encode(state, m.model)
+  },
+  decode(state) {
+    const r0 = c.string.decode(state)
+
+    return {
+      model: r0
+    }
+  }
+}
+
+// @harness/unload-model-response
+const encoding91 = {
+  preencode(state, m) {
+    state.end++ // max flag is 1 so always one byte
+  },
+  encode(state, m) {
+    const flags = (version >= 16 && m.unloaded) ? 1 : 0
+
+    c.uint.encode(state, flags)
+  },
+  decode(state) {
+    const flags = c.uint.decode(state)
+
+    return {
+      unloaded: (version >= 16 && (flags & 1) !== 0)
+    }
+  }
+}
+
 // @harness/completion-frame.grantedDrives, deferred due to recusive use
 const encoding11_23 = c.array(c.frame(encoding60))
 // @harness/skill.setupV2, deferred due to recusive use
@@ -3605,6 +3665,12 @@ function getEncoding(name) {
       return encoding87
     case '@harness/set-model-idle-timeout-request':
       return encoding88
+    case '@harness/load-model-response':
+      return encoding89
+    case '@harness/unload-model-request':
+      return encoding90
+    case '@harness/unload-model-response':
+      return encoding91
     default:
       throw new Error('Encoder not found ' + name)
   }

package/dist-lib/spec/hyperschema/schema.json

@@ -1,5 +1,5 @@
 {
-  "version": 15,
+  "version": 16,
   "schema": [
     {
       "name": "empty",
@@ -2717,6 +2717,53 @@
           "version": 15
         }
       ]
+    },
+    {
+      "name": "load-model-response",
+      "namespace": "harness",
+      "compact": false,
+      "flagsPosition": 0,
+      "fields": [
+        {
+          "name": "loaded",
+          "required": false,
+          "type": "bool",
+          "version": 16
+        },
+        {
+          "name": "reason",
+          "type": "string",
+          "version": 16
+        }
+      ]
+    },
+    {
+      "name": "unload-model-request",
+      "namespace": "harness",
+      "compact": false,
+      "flagsPosition": -1,
+      "fields": [
+        {
+          "name": "model",
+          "required": true,
+          "type": "string",
+          "version": 16
+        }
+      ]
+    },
+    {
+      "name": "unload-model-response",
+      "namespace": "harness",
+      "compact": false,
+      "flagsPosition": 0,
+      "fields": [
+        {
+          "name": "unloaded",
+          "required": false,
+          "type": "bool",
+          "version": 16
+        }
+      ]
     }
   ]
 }

package/dist-lib/spec/hyperschema/types.d.ts

@@ -1,5 +1,5 @@
 // This file is autogenerated by hyperschema-ts
-// Schema version: 15
+// Schema version: 16
 // Do not edit manually.
 
 export interface HarnessEmpty {
@@ -568,6 +568,19 @@ export interface HarnessSetModelIdleTimeoutRequest {
   ms: number
 }
 
+export interface HarnessLoadModelResponse {
+  loaded?: boolean
+  reason?: string | null
+}
+
+export interface HarnessUnloadModelRequest {
+  model: string
+}
+
+export interface HarnessUnloadModelResponse {
+  unloaded?: boolean
+}
+
 export interface SchemaTypes {
   "@harness/empty": HarnessEmpty
   "@harness/ok-response": HarnessOkResponse
@@ -658,4 +671,7 @@ export interface SchemaTypes {
   "@harness/model-source-file": HarnessModelSourceFile
   "@harness/preview-model-source-response": HarnessPreviewModelSourceResponse
   "@harness/set-model-idle-timeout-request": HarnessSetModelIdleTimeoutRequest
+  "@harness/load-model-response": HarnessLoadModelResponse
+  "@harness/unload-model-request": HarnessUnloadModelRequest
+  "@harness/unload-model-response": HarnessUnloadModelResponse
 }

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@qvac/harness",
-  "version": "0.5.0",
+  "version": "0.6.0",
   "type": "module",
   "description": "An AI harness implementing generate(context) via @qvac/sdk. Knows nothing about workbench, hyperdb, or chunks — just inference in, events out.",
   "exports": {
@@ -606,6 +606,7 @@
     "build:bundle": "node scripts/build-bundle.mjs",
     "skill-eval": "bare examples/skill-eval/run.ts",
     "translate-eval": "bare examples/translate-eval/run.ts",
+    "model-residency": "bare examples/model-residency/run.ts",
     "prepack": "node scripts/vendor-python-wheels.mjs && node scripts/build-lib.mjs && node scripts/build-bundle.mjs --minify",
     "vendor:wheels": "node scripts/vendor-python-wheels.mjs",
     "build:lib": "node scripts/build-lib.mjs",