chore: deploy Laya Tetris c2eb1a8

deploy-sha.txt

@@ -1 +1 @@
-523ab63a3b9bacfaf931fdcee52e061a3ffc8c58
+c2eb1a822d9dfd9cb902afe789dabe69e451ad96

flutter_bootstrap.js

@@ -42,6 +42,6 @@ _flutter.buildConfig = {"engineRevision":"5d531788691ec3404cac0cee66ead4007b1773
 
 _flutter.loader.load({
   serviceWorkerSettings: {
-    serviceWorkerVersion: "3544709619" /* Flutter's service worker is deprecated and will be removed in a future Flutter release. */
+    serviceWorkerVersion: "1346589549" /* Flutter's service worker is deprecated and will be removed in a future Flutter release. */
   }
 });

webgpu_bridge/llama_webgpu_bridge.d.ts

@@ -241,7 +241,7 @@ export type GrammarTrigger =
   | { type: 'word' | 'pattern' | 'pattern-full'; value: string }
   | { type: 'token'; token: number };
 
-/** llama.cpp's reasoning budget: after `maxTokens` tokens inside a reasoning block, plus any that finish a UTF-8 character, the sampler forces `forcedMessage` and `endTag`. A grammar pauses after `startTag` until `endTag` completes. */
+/** llama.cpp's reasoning budget: after `maxTokens` tokens inside a reasoning block, plus any that finish a UTF-8 character, the sampler forces `forcedMessage` and `endTag`. A grammar pauses after `startTag` until `endTag` completes, unless `pauseGrammar` is false. */
 export interface ThinkingBudgetOptions {
   /** Integer from 0 to 2147483647. 0 forces the end as soon as a block opens, or at once when the prompt ends inside one. */
   maxTokens: number;
@@ -251,6 +251,8 @@ export interface ThinkingBudgetOptions {
   endTag: string;
   /** Text forced before `endTag` when the budget runs out; empty by default. */
   forcedMessage?: string;
+  /** `false` keeps `grammar` applied inside a reasoning block, for a grammar that takes the block and `endTag` itself, such as one written for a prompt that ends inside the block; paused, it would resume at its start and take the reply as reasoning. It must accept `forcedMessage` and `endTag` wherever the budget can run out. Omitted, null or true pauses the grammar. Needs the `thinkingBudgetPauseGrammar` capability. */
+  pauseGrammar?: boolean | null;
 }
 
 /** Completion options the loaded core applies. Every flag is false until a model load initializes the core. */
@@ -261,6 +263,8 @@ export interface CompletionCapabilities {
   minP: boolean;
   /** `CompletionOptions.thinkingBudget` is applied. */
   thinkingBudget: boolean;
+  /** `CompletionOptions.thinkingBudget.pauseGrammar` is applied. A core without it pauses every grammar inside a reasoning block. */
+  thinkingBudgetPauseGrammar: boolean;
   /** The penalties count the prompt, and `CompletionOptions.penaltyPromptTokens` is applied. A core without it never counts the prompt, so `penaltyPromptTokens: false` holds there too. */
   penaltyPromptTokens: boolean;
   /** The loaded core applies CompletionOptions.penaltyPromptTokensWithGrammar. Only literal true permits the opt-in. */

webgpu_bridge/llama_webgpu_bridge.js

@@ -239,6 +239,7 @@ var NO_COMPLETION_CAPABILITIES = Object.freeze({
   presencePenalty: false,
   minP: false,
   thinkingBudget: false,
+  thinkingBudgetPauseGrammar: false,
   penaltyPromptTokens: false,
   penaltyPromptTokensWithGrammar: false,
   lazyGrammar: false,
@@ -284,10 +285,16 @@ function thinkingBudget(value, hasMediaParts) {
     throw new TypeError("CompletionOptions.thinkingBudget.forcedMessage must be a string.");
   }
   withoutNul(forcedMessage, "CompletionOptions.thinkingBudget.forcedMessage");
+  const pauseGrammar = budget.pauseGrammar ?? true;
+  if (typeof pauseGrammar !== "boolean") {
+    throw new TypeError(
+      `CompletionOptions.thinkingBudget.pauseGrammar must be a boolean; got ${String(pauseGrammar)}.`
+    );
+  }
   if (hasMediaParts) {
     throw new Error("CompletionOptions.thinkingBudget supports text-only prompts; remove the media parts.");
   }
-  return { maxTokens, startTag, endTag, forcedMessage };
+  return { maxTokens, startTag, endTag, forcedMessage, pauseGrammar };
 }
 function speculativeInteger(config, name, min, max) {
   const value = config[name];
@@ -545,6 +552,7 @@ function completionCapabilitiesFrom(raw) {
     presencePenalty: parsed?.presencePenalty === true,
     minP: parsed?.minP === true,
     thinkingBudget: parsed?.thinkingBudget === true,
+    thinkingBudgetPauseGrammar: parsed?.thinkingBudgetPauseGrammar === true,
     penaltyPromptTokens: parsed?.penaltyPromptTokens === true,
     penaltyPromptTokensWithGrammar: parsed?.penaltyPromptTokensWithGrammar === true,
     lazyGrammar: parsed?.lazyGrammar === true,
@@ -557,6 +565,7 @@ function requireCompletionCapabilities(sampling, capabilities) {
     sampling.minP !== 0 && !capabilities.minP ? "minP" : null,
     sampling.presencePenalty !== 0 && !capabilities.presencePenalty ? "presencePenalty" : null,
     sampling.thinkingBudget != null && !capabilities.thinkingBudget ? "thinkingBudget" : null,
+    sampling.thinkingBudget?.pauseGrammar === false && !capabilities.thinkingBudgetPauseGrammar ? "thinkingBudget.pauseGrammar" : null,
     sampling.grammarOptions?.lazy === true && !capabilities.lazyGrammar ? "grammarLazy" : null,
     sampling.grammarOptions != null && sampling.grammarOptions.root !== "root" && !capabilities.lazyGrammar ? "grammarRoot" : null,
     sampling.speculative != null && !Object.values(capabilities.speculativeDecoding).includes(true) ? "speculativeDecoding" : null
@@ -4570,6 +4579,9 @@ var LlamaWebGpuBridgeRuntime = class {
       if (sampling.grammarOptions) {
         this._setNextGrammar(sampling.grammarOptions);
       }
+      if (sampling.thinkingBudget?.pauseGrammar === false) {
+        this._core.ccall("llamadart_webgpu_set_next_grammar_reasoning_pause", null, ["number"], [0]);
+      }
       const beginRc = Number(
         await this._core.ccall(
           input.exportName,
@@ -6885,7 +6897,7 @@ var LlamaWebGpuBridge = class {
   }
   async _createCompletionUnlocked(prompt, options = {}) {
     const sampling = resolveCompletionSamplingOptions(options);
-    if (sampling.penaltyPromptTokensWithGrammar) {
+    if (sampling.penaltyPromptTokensWithGrammar || sampling.thinkingBudget?.pauseGrammar === false) {
       requireCompletionCapabilities(sampling, await this._getCompletionCapabilitiesUnlocked());
     }
     const isWarmup = options?.warmup === true;

webgpu_bridge/llama_webgpu_core.wasm

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f20057d88b90de59da5ad35adb8aa49fa4ef5156c9e64e2957d25f8cafe38bdd
-size 11099140
+oid sha256:7e17fac640d50a33a2277036a891b2939f023c4864652b9866236092ac2cb6ee
+size 11099356

webgpu_bridge/llama_webgpu_core_mem64.wasm

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c2d6548505893b6d89e2a61d6b00a2c502ff03ac8e691c6328c048bbdd4f643a
-size 11390403
+oid sha256:249ab51961555ccf42139b7cb1c05cbd83f62ce3d181aad45671cfc83bb94501
+size 11390643

webgpu_bridge/manifest.json

@@ -1,37 +1,37 @@
 {
   "artifacts": {
     "llama_webgpu_bridge.d.ts": {
-      "sha256": "8a5095c3b5fc62f10612cc6cf3326bc3065529103ad657857f4e1a9fe58dc6ed",
-      "size_bytes": 22479
+      "sha256": "766adcfe0b48c07c26c478a7c9b9dc72bff9386ff297c532baae709ff6367db4",
+      "size_bytes": 23141
     },
     "llama_webgpu_bridge.js": {
-      "sha256": "eaefb36f0d535ac7f76cadecfe97c66c416fc49cb05a4f87144f84499368d79c",
-      "size_bytes": 316712
+      "sha256": "0ecc0ebb7a0f752f48a95de7cefe1e8f2f6b4e955e8a4674f09d49e03dbcb06e",
+      "size_bytes": 317431
     },
     "llama_webgpu_bridge_worker.js": {
       "sha256": "47bbfa0fe897e708455b9497e9aa41d2e94489c329300c33b12936fb3169deca",
       "size_bytes": 257
     },
     "llama_webgpu_core.js": {
-      "sha256": "58b74b403eabfde0703e1f53cdbb9fd34e5eee480bcd0443b32cf97ba48448c0",
-      "size_bytes": 123402
+      "sha256": "f34ad08e95cf6dd1bddd7df31794383ceb6e350553f4487746f54210f8fd9c18",
+      "size_bytes": 123583
     },
     "llama_webgpu_core.wasm": {
-      "sha256": "f20057d88b90de59da5ad35adb8aa49fa4ef5156c9e64e2957d25f8cafe38bdd",
-      "size_bytes": 11099140
+      "sha256": "7e17fac640d50a33a2277036a891b2939f023c4864652b9866236092ac2cb6ee",
+      "size_bytes": 11099356
     },
     "llama_webgpu_core_mem64.js": {
-      "sha256": "282d0934ec2090b13f527e1f01d5740374bdfcc1e75c1f46f9148e7007976a0b",
-      "size_bytes": 147159
+      "sha256": "32b4689944aa4e3a940bddaf31928ab3476e57b9963b918cb38c1281e5688297",
+      "size_bytes": 147340
     },
     "llama_webgpu_core_mem64.wasm": {
-      "sha256": "c2d6548505893b6d89e2a61d6b00a2c502ff03ac8e691c6328c048bbdd4f643a",
-      "size_bytes": 11390403
+      "sha256": "249ab51961555ccf42139b7cb1c05cbd83f62ce3d181aad45671cfc83bb94501",
+      "size_bytes": 11390643
     }
   },
   "assets_repository": "leehack/llama-web-bridge-assets",
-  "bridge_assets_tag": "v0.1.83",
-  "bridge_commit": "cb6383fbe7e903764a89f5c80c7f81c219f99769",
+  "bridge_assets_tag": "v0.1.84",
+  "bridge_commit": "dc359398bc6445ec15878ddb873555295f957f85",
   "bridge_repository": "leehack/llama-web-bridge",
   "capabilities": {
     "memory64": true,
@@ -62,43 +62,43 @@
   "emscripten_version": "6.0.8",
   "files": {
     "llama_webgpu_bridge.d.ts": {
-      "sha256": "8a5095c3b5fc62f10612cc6cf3326bc3065529103ad657857f4e1a9fe58dc6ed",
-      "size_bytes": 22479
+      "sha256": "766adcfe0b48c07c26c478a7c9b9dc72bff9386ff297c532baae709ff6367db4",
+      "size_bytes": 23141
     },
     "llama_webgpu_bridge.js": {
-      "sha256": "eaefb36f0d535ac7f76cadecfe97c66c416fc49cb05a4f87144f84499368d79c",
-      "size_bytes": 316712
+      "sha256": "0ecc0ebb7a0f752f48a95de7cefe1e8f2f6b4e955e8a4674f09d49e03dbcb06e",
+      "size_bytes": 317431
     },
     "llama_webgpu_bridge_worker.js": {
       "sha256": "47bbfa0fe897e708455b9497e9aa41d2e94489c329300c33b12936fb3169deca",
       "size_bytes": 257
     },
     "llama_webgpu_core.js": {
-      "sha256": "58b74b403eabfde0703e1f53cdbb9fd34e5eee480bcd0443b32cf97ba48448c0",
-      "size_bytes": 123402
+      "sha256": "f34ad08e95cf6dd1bddd7df31794383ceb6e350553f4487746f54210f8fd9c18",
+      "size_bytes": 123583
     },
     "llama_webgpu_core.wasm": {
-      "sha256": "f20057d88b90de59da5ad35adb8aa49fa4ef5156c9e64e2957d25f8cafe38bdd",
-      "size_bytes": 11099140
+      "sha256": "7e17fac640d50a33a2277036a891b2939f023c4864652b9866236092ac2cb6ee",
+      "size_bytes": 11099356
     },
     "llama_webgpu_core_mem64.js": {
-      "sha256": "282d0934ec2090b13f527e1f01d5740374bdfcc1e75c1f46f9148e7007976a0b",
-      "size_bytes": 147159
+      "sha256": "32b4689944aa4e3a940bddaf31928ab3476e57b9963b918cb38c1281e5688297",
+      "size_bytes": 147340
     },
     "llama_webgpu_core_mem64.wasm": {
-      "sha256": "c2d6548505893b6d89e2a61d6b00a2c502ff03ac8e691c6328c048bbdd4f643a",
-      "size_bytes": 11390403
+      "sha256": "249ab51961555ccf42139b7cb1c05cbd83f62ce3d181aad45671cfc83bb94501",
+      "size_bytes": 11390643
     }
   },
-  "github_run_id": "42479331139",
-  "github_run_url": "https://github.com/leehack/llama-web-bridge/actions/runs/42479331139",
+  "github_run_id": "43272135497",
+  "github_run_url": "https://github.com/leehack/llama-web-bridge/actions/runs/43272135497",
   "llama_cpp_commit": "cd959bae99ff4bed22bef2604f45241c09bbb8c8",
   "llama_cpp_tag": "v0.9.0",
   "native_commit": "a604639131105c42a25a3e7384ca10314d385f80",
   "native_manifest_sha256": "0d2119cdcf34d347a96a996423094498de3393026efb6ad1a96ba286b621c39e",
   "native_release_tag": "v0.9.0",
   "native_repository": "leehack/llamadart-native",
-  "orchestrator_correlation_id": "manual-v0.1.83",
+  "orchestrator_correlation_id": "manual-v0.1.84",
   "qualification_gates": {
     "multimodal": "passed",
     "speech_to_text": "required-automated-qualification",
@@ -107,9 +107,9 @@
   },
   "release_channel": "stable",
   "release_rebuild": 0,
-  "release_tag": "v0.1.83",
+  "release_tag": "v0.1.84",
   "schema_version": 2,
-  "source_commit": "cb6383fbe7e903764a89f5c80c7f81c219f99769",
+  "source_commit": "dc359398bc6445ec15878ddb873555295f957f85",
   "source_repository": "leehack/llama-web-bridge",
   "unproven_capabilities": {
     "hardware_gpu_acceleration": "unavailable-on-hosted-runners",

webgpu_bridge/sha256sums.txt

@@ -1,7 +1,7 @@
-eaefb36f0d535ac7f76cadecfe97c66c416fc49cb05a4f87144f84499368d79c  llama_webgpu_bridge.js
-58b74b403eabfde0703e1f53cdbb9fd34e5eee480bcd0443b32cf97ba48448c0  llama_webgpu_core.js
-f20057d88b90de59da5ad35adb8aa49fa4ef5156c9e64e2957d25f8cafe38bdd  llama_webgpu_core.wasm
+0ecc0ebb7a0f752f48a95de7cefe1e8f2f6b4e955e8a4674f09d49e03dbcb06e  llama_webgpu_bridge.js
+f34ad08e95cf6dd1bddd7df31794383ceb6e350553f4487746f54210f8fd9c18  llama_webgpu_core.js
+7e17fac640d50a33a2277036a891b2939f023c4864652b9866236092ac2cb6ee  llama_webgpu_core.wasm
 47bbfa0fe897e708455b9497e9aa41d2e94489c329300c33b12936fb3169deca  llama_webgpu_bridge_worker.js
-8a5095c3b5fc62f10612cc6cf3326bc3065529103ad657857f4e1a9fe58dc6ed  llama_webgpu_bridge.d.ts
-282d0934ec2090b13f527e1f01d5740374bdfcc1e75c1f46f9148e7007976a0b  llama_webgpu_core_mem64.js
-c2d6548505893b6d89e2a61d6b00a2c502ff03ac8e691c6328c048bbdd4f643a  llama_webgpu_core_mem64.wasm
+766adcfe0b48c07c26c478a7c9b9dc72bff9386ff297c532baae709ff6367db4  llama_webgpu_bridge.d.ts
+32b4689944aa4e3a940bddaf31928ab3476e57b9963b918cb38c1281e5688297  llama_webgpu_core_mem64.js
+249ab51961555ccf42139b7cb1c05cbd83f62ce3d181aad45671cfc83bb94501  llama_webgpu_core_mem64.wasm