Spaces:
Running
Running
chore: deploy Laya Tetris 47dfa73
Browse files- deploy-sha.txt +1 -1
- flutter_bootstrap.js +1 -1
- webgpu_bridge/llama_webgpu_bridge.js +40 -4
- webgpu_bridge/manifest.json +11 -11
- webgpu_bridge/sha256sums.txt +1 -1
deploy-sha.txt
CHANGED
|
@@ -1 +1 @@
|
|
| 1 |
-
|
|
|
|
| 1 |
+
47dfa7369f4f1952f3b5425b6a0a0ce33d0b7bb8
|
flutter_bootstrap.js
CHANGED
|
@@ -42,6 +42,6 @@ _flutter.buildConfig = {"engineRevision":"5d531788691ec3404cac0cee66ead4007b1773
|
|
| 42 |
|
| 43 |
_flutter.loader.load({
|
| 44 |
serviceWorkerSettings: {
|
| 45 |
-
serviceWorkerVersion: "
|
| 46 |
}
|
| 47 |
});
|
|
|
|
| 42 |
|
| 43 |
_flutter.loader.load({
|
| 44 |
serviceWorkerSettings: {
|
| 45 |
+
serviceWorkerVersion: "2520760172" /* Flutter's service worker is deprecated and will be removed in a future Flutter release. */
|
| 46 |
}
|
| 47 |
});
|
webgpu_bridge/llama_webgpu_bridge.js
CHANGED
|
@@ -1484,6 +1484,7 @@ var LlamaWebGpuBridgeRuntime = class {
|
|
| 1484 |
const previousPreferMemory64 = this._preferMemory64;
|
| 1485 |
const previousMMProjSourceUrl = this._mmProjSourceUrl;
|
| 1486 |
const previousDraftModel = this._draftModel;
|
|
|
|
| 1487 |
try {
|
| 1488 |
this._preferMemory64 = false;
|
| 1489 |
await this.loadModelFromUrl(modelUrl, {
|
|
@@ -1495,7 +1496,7 @@ var LlamaWebGpuBridgeRuntime = class {
|
|
| 1495 |
remoteFetchChunkBytes: this._resolveRemoteFetchChunkBytes(options),
|
| 1496 |
safariGpuProbe: false,
|
| 1497 |
signal: options.signal
|
| 1498 |
-
});
|
| 1499 |
if (typeof previousMMProjSourceUrl === "string" && previousMMProjSourceUrl.length > 0) {
|
| 1500 |
try {
|
| 1501 |
await this.loadMultimodalProjector(previousMMProjSourceUrl);
|
|
@@ -1790,6 +1791,28 @@ var LlamaWebGpuBridgeRuntime = class {
|
|
| 1790 |
_nativeLoadOptionTypes() {
|
| 1791 |
return this._nativeLoadOptionValues().map(() => "number");
|
| 1792 |
}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1793 |
async _tryLoadModelFromRemoteFetchBackend(core, url, options = {}) {
|
| 1794 |
if (!this._canUseRemoteFetchBackend(options)) {
|
| 1795 |
return { loaded: false, sizeBytes: null };
|
|
@@ -2546,7 +2569,7 @@ var LlamaWebGpuBridgeRuntime = class {
|
|
| 2546 |
}
|
| 2547 |
return this._gpuActive;
|
| 2548 |
}
|
| 2549 |
-
async loadModelFromUrl(url, options = {}) {
|
| 2550 |
const operationSignal = options?.signal || null;
|
| 2551 |
const abortMessage = "Model load was cancelled.";
|
| 2552 |
throwIfAborted(operationSignal, abortMessage);
|
|
@@ -2602,7 +2625,13 @@ var LlamaWebGpuBridgeRuntime = class {
|
|
| 2602 |
if (this._nUbatch > 0 && this._nBatch > 0 && this._nUbatch > this._nBatch) {
|
| 2603 |
this._nUbatch = this._nBatch;
|
| 2604 |
}
|
| 2605 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 2606 |
if (Number.isFinite(this._threadPoolSizeHint) && this._threadPoolSizeHint > 0) {
|
| 2607 |
this._pushRuntimeNote(`thread_pool_size:${this._threadPoolSizeHint}`);
|
| 2608 |
}
|
|
@@ -2625,7 +2654,7 @@ var LlamaWebGpuBridgeRuntime = class {
|
|
| 2625 |
if (this._nSeqMax > 0) {
|
| 2626 |
this._pushRuntimeNote(`n_seq_max:${this._nSeqMax}`);
|
| 2627 |
}
|
| 2628 |
-
if (isCpuModelMode && !Number.isFinite(requestedBatch) && !Number.isFinite(requestedUbatch)) {
|
| 2629 |
this._runtimeNotes.push("cpu_batch_tuned_default");
|
| 2630 |
}
|
| 2631 |
if (this._isSafari && this._nGpuLayers > 0) {
|
|
@@ -5962,6 +5991,13 @@ var LlamaWebGpuBridge = class {
|
|
| 5962 |
_workerConfig() {
|
| 5963 |
const config = this._config || {};
|
| 5964 |
return {
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 5965 |
wasmUrl: typeof config.wasmUrl === "string" ? config.wasmUrl : void 0,
|
| 5966 |
wasmUrlMem64: typeof config.wasmUrlMem64 === "string" ? config.wasmUrlMem64 : void 0,
|
| 5967 |
coreModuleUrl: typeof config.coreModuleUrl === "string" ? config.coreModuleUrl : void 0,
|
|
|
|
| 1484 |
const previousPreferMemory64 = this._preferMemory64;
|
| 1485 |
const previousMMProjSourceUrl = this._mmProjSourceUrl;
|
| 1486 |
const previousDraftModel = this._draftModel;
|
| 1487 |
+
const resolvedLoadState = this._resolvedLoadState();
|
| 1488 |
try {
|
| 1489 |
this._preferMemory64 = false;
|
| 1490 |
await this.loadModelFromUrl(modelUrl, {
|
|
|
|
| 1496 |
remoteFetchChunkBytes: this._resolveRemoteFetchChunkBytes(options),
|
| 1497 |
safariGpuProbe: false,
|
| 1498 |
signal: options.signal
|
| 1499 |
+
}, resolvedLoadState);
|
| 1500 |
if (typeof previousMMProjSourceUrl === "string" && previousMMProjSourceUrl.length > 0) {
|
| 1501 |
try {
|
| 1502 |
await this.loadMultimodalProjector(previousMMProjSourceUrl);
|
|
|
|
| 1791 |
_nativeLoadOptionTypes() {
|
| 1792 |
return this._nativeLoadOptionValues().map(() => "number");
|
| 1793 |
}
|
| 1794 |
+
_resolvedLoadState() {
|
| 1795 |
+
return {
|
| 1796 |
+
_nCtx: this._nCtx,
|
| 1797 |
+
_threads: this._threads,
|
| 1798 |
+
_threadsBatch: this._threadsBatch,
|
| 1799 |
+
_nBatch: this._nBatch,
|
| 1800 |
+
_nUbatch: this._nUbatch,
|
| 1801 |
+
_nSeqMax: this._nSeqMax,
|
| 1802 |
+
_useMmap: this._useMmap,
|
| 1803 |
+
_useMlock: this._useMlock,
|
| 1804 |
+
_flashAttention: this._flashAttention,
|
| 1805 |
+
_cacheTypeK: this._cacheTypeK,
|
| 1806 |
+
_cacheTypeV: this._cacheTypeV,
|
| 1807 |
+
_kvUnified: this._kvUnified,
|
| 1808 |
+
_ropeFrequencyBase: this._ropeFrequencyBase,
|
| 1809 |
+
_ropeFrequencyScale: this._ropeFrequencyScale,
|
| 1810 |
+
_splitMode: this._splitMode,
|
| 1811 |
+
_mainGpu: this._mainGpu,
|
| 1812 |
+
_loadMtp: this._loadMtp,
|
| 1813 |
+
_speculativeRollbackTokenMax: this._speculativeRollbackTokenMax
|
| 1814 |
+
};
|
| 1815 |
+
}
|
| 1816 |
async _tryLoadModelFromRemoteFetchBackend(core, url, options = {}) {
|
| 1817 |
if (!this._canUseRemoteFetchBackend(options)) {
|
| 1818 |
return { loaded: false, sizeBytes: null };
|
|
|
|
| 2569 |
}
|
| 2570 |
return this._gpuActive;
|
| 2571 |
}
|
| 2572 |
+
async loadModelFromUrl(url, options = {}, resolvedLoadState = null) {
|
| 2573 |
const operationSignal = options?.signal || null;
|
| 2574 |
const abortMessage = "Model load was cancelled.";
|
| 2575 |
throwIfAborted(operationSignal, abortMessage);
|
|
|
|
| 2625 |
if (this._nUbatch > 0 && this._nBatch > 0 && this._nUbatch > this._nBatch) {
|
| 2626 |
this._nUbatch = this._nBatch;
|
| 2627 |
}
|
| 2628 |
+
if (resolvedLoadState) {
|
| 2629 |
+
Object.assign(this, resolvedLoadState);
|
| 2630 |
+
this._threads = this._capThreadsToPool(this._threads);
|
| 2631 |
+
this._threadsBatch = this._capThreadsToPool(this._threadsBatch, { noteTag: "threads_batch_capped_pool" });
|
| 2632 |
+
} else {
|
| 2633 |
+
this._resolveNativeLoadOptions(options);
|
| 2634 |
+
}
|
| 2635 |
if (Number.isFinite(this._threadPoolSizeHint) && this._threadPoolSizeHint > 0) {
|
| 2636 |
this._pushRuntimeNote(`thread_pool_size:${this._threadPoolSizeHint}`);
|
| 2637 |
}
|
|
|
|
| 2654 |
if (this._nSeqMax > 0) {
|
| 2655 |
this._pushRuntimeNote(`n_seq_max:${this._nSeqMax}`);
|
| 2656 |
}
|
| 2657 |
+
if (!resolvedLoadState && isCpuModelMode && !Number.isFinite(requestedBatch) && !Number.isFinite(requestedUbatch)) {
|
| 2658 |
this._runtimeNotes.push("cpu_batch_tuned_default");
|
| 2659 |
}
|
| 2660 |
if (this._isSafari && this._nGpuLayers > 0) {
|
|
|
|
| 5991 |
_workerConfig() {
|
| 5992 |
const config = this._config || {};
|
| 5993 |
return {
|
| 5994 |
+
workerRequestTimeoutMs: config.workerRequestTimeoutMs,
|
| 5995 |
+
workerInitTimeoutMs: config.workerInitTimeoutMs,
|
| 5996 |
+
workerModelLoadTimeoutMs: config.workerModelLoadTimeoutMs,
|
| 5997 |
+
workerMmprojLoadTimeoutMs: config.workerMmprojLoadTimeoutMs,
|
| 5998 |
+
workerCompletionTimeoutMs: config.workerCompletionTimeoutMs,
|
| 5999 |
+
workerTextToSpeechTimeoutMs: config.workerTextToSpeechTimeoutMs,
|
| 6000 |
+
coreInitTimeoutMs: config.coreInitTimeoutMs,
|
| 6001 |
wasmUrl: typeof config.wasmUrl === "string" ? config.wasmUrl : void 0,
|
| 6002 |
wasmUrlMem64: typeof config.wasmUrlMem64 === "string" ? config.wasmUrlMem64 : void 0,
|
| 6003 |
coreModuleUrl: typeof config.coreModuleUrl === "string" ? config.coreModuleUrl : void 0,
|
webgpu_bridge/manifest.json
CHANGED
|
@@ -5,8 +5,8 @@
|
|
| 5 |
"size_bytes": 19006
|
| 6 |
},
|
| 7 |
"llama_webgpu_bridge.js": {
|
| 8 |
-
"sha256": "
|
| 9 |
-
"size_bytes":
|
| 10 |
},
|
| 11 |
"llama_webgpu_bridge_worker.js": {
|
| 12 |
"sha256": "47bbfa0fe897e708455b9497e9aa41d2e94489c329300c33b12936fb3169deca",
|
|
@@ -30,8 +30,8 @@
|
|
| 30 |
}
|
| 31 |
},
|
| 32 |
"assets_repository": "leehack/llama-web-bridge-assets",
|
| 33 |
-
"bridge_assets_tag": "v0.1.
|
| 34 |
-
"bridge_commit": "
|
| 35 |
"bridge_repository": "leehack/llama-web-bridge",
|
| 36 |
"capabilities": {
|
| 37 |
"memory64": true,
|
|
@@ -66,8 +66,8 @@
|
|
| 66 |
"size_bytes": 19006
|
| 67 |
},
|
| 68 |
"llama_webgpu_bridge.js": {
|
| 69 |
-
"sha256": "
|
| 70 |
-
"size_bytes":
|
| 71 |
},
|
| 72 |
"llama_webgpu_bridge_worker.js": {
|
| 73 |
"sha256": "47bbfa0fe897e708455b9497e9aa41d2e94489c329300c33b12936fb3169deca",
|
|
@@ -90,15 +90,15 @@
|
|
| 90 |
"size_bytes": 10461625
|
| 91 |
}
|
| 92 |
},
|
| 93 |
-
"github_run_id": "
|
| 94 |
-
"github_run_url": "https://github.com/leehack/llama-web-bridge/actions/runs/
|
| 95 |
"llama_cpp_commit": "620d2225cc436ab0b001e3200e590e3b286ef9ab",
|
| 96 |
"llama_cpp_tag": "v0.7.0",
|
| 97 |
"native_commit": "daffc7fc8bdb62bbc09f723a74e761f6f32d549f",
|
| 98 |
"native_manifest_sha256": "ddcd7aecdbf129421b02a35224e6632059f97523baa0f214ef369710ae8446ce",
|
| 99 |
"native_release_tag": "v0.7.0-1",
|
| 100 |
"native_repository": "leehack/llamadart-native",
|
| 101 |
-
"orchestrator_correlation_id": "
|
| 102 |
"qualification_gates": {
|
| 103 |
"multimodal": "passed",
|
| 104 |
"speech_to_text": "required-automated-qualification",
|
|
@@ -107,9 +107,9 @@
|
|
| 107 |
},
|
| 108 |
"release_channel": "stable",
|
| 109 |
"release_rebuild": 0,
|
| 110 |
-
"release_tag": "v0.1.
|
| 111 |
"schema_version": 2,
|
| 112 |
-
"source_commit": "
|
| 113 |
"source_repository": "leehack/llama-web-bridge",
|
| 114 |
"unproven_capabilities": {
|
| 115 |
"hardware_gpu_acceleration": "unavailable-on-hosted-runners",
|
|
|
|
| 5 |
"size_bytes": 19006
|
| 6 |
},
|
| 7 |
"llama_webgpu_bridge.js": {
|
| 8 |
+
"sha256": "f4568c8074cfd439c3fe33741e69bf01820149ea5d63e477db703fc7eaa13e8f",
|
| 9 |
+
"size_bytes": 295516
|
| 10 |
},
|
| 11 |
"llama_webgpu_bridge_worker.js": {
|
| 12 |
"sha256": "47bbfa0fe897e708455b9497e9aa41d2e94489c329300c33b12936fb3169deca",
|
|
|
|
| 30 |
}
|
| 31 |
},
|
| 32 |
"assets_repository": "leehack/llama-web-bridge-assets",
|
| 33 |
+
"bridge_assets_tag": "v0.1.66",
|
| 34 |
+
"bridge_commit": "de2c3727c4b8ae9dad9b887bd89f1cb43793f008",
|
| 35 |
"bridge_repository": "leehack/llama-web-bridge",
|
| 36 |
"capabilities": {
|
| 37 |
"memory64": true,
|
|
|
|
| 66 |
"size_bytes": 19006
|
| 67 |
},
|
| 68 |
"llama_webgpu_bridge.js": {
|
| 69 |
+
"sha256": "f4568c8074cfd439c3fe33741e69bf01820149ea5d63e477db703fc7eaa13e8f",
|
| 70 |
+
"size_bytes": 295516
|
| 71 |
},
|
| 72 |
"llama_webgpu_bridge_worker.js": {
|
| 73 |
"sha256": "47bbfa0fe897e708455b9497e9aa41d2e94489c329300c33b12936fb3169deca",
|
|
|
|
| 90 |
"size_bytes": 10461625
|
| 91 |
}
|
| 92 |
},
|
| 93 |
+
"github_run_id": "39139757535",
|
| 94 |
+
"github_run_url": "https://github.com/leehack/llama-web-bridge/actions/runs/39139757535",
|
| 95 |
"llama_cpp_commit": "620d2225cc436ab0b001e3200e590e3b286ef9ab",
|
| 96 |
"llama_cpp_tag": "v0.7.0",
|
| 97 |
"native_commit": "daffc7fc8bdb62bbc09f723a74e761f6f32d549f",
|
| 98 |
"native_manifest_sha256": "ddcd7aecdbf129421b02a35224e6632059f97523baa0f214ef369710ae8446ce",
|
| 99 |
"native_release_tag": "v0.7.0-1",
|
| 100 |
"native_repository": "leehack/llamadart-native",
|
| 101 |
+
"orchestrator_correlation_id": "qualification-only-20261018-v0166-de2c372",
|
| 102 |
"qualification_gates": {
|
| 103 |
"multimodal": "passed",
|
| 104 |
"speech_to_text": "required-automated-qualification",
|
|
|
|
| 107 |
},
|
| 108 |
"release_channel": "stable",
|
| 109 |
"release_rebuild": 0,
|
| 110 |
+
"release_tag": "v0.1.66",
|
| 111 |
"schema_version": 2,
|
| 112 |
+
"source_commit": "de2c3727c4b8ae9dad9b887bd89f1cb43793f008",
|
| 113 |
"source_repository": "leehack/llama-web-bridge",
|
| 114 |
"unproven_capabilities": {
|
| 115 |
"hardware_gpu_acceleration": "unavailable-on-hosted-runners",
|
webgpu_bridge/sha256sums.txt
CHANGED
|
@@ -1,4 +1,4 @@
|
|
| 1 |
-
|
| 2 |
ba73744e5ae40e4eea8e6d4f1e2db5c36d1b983b03eb8fa883f4b9969bbcfe5e llama_webgpu_core.js
|
| 3 |
5be341eeb6fbf19f731fde155bb3afc71cf8553e0a8e6fc9ce10ad5917e792e5 llama_webgpu_core.wasm
|
| 4 |
47bbfa0fe897e708455b9497e9aa41d2e94489c329300c33b12936fb3169deca llama_webgpu_bridge_worker.js
|
|
|
|
| 1 |
+
f4568c8074cfd439c3fe33741e69bf01820149ea5d63e477db703fc7eaa13e8f llama_webgpu_bridge.js
|
| 2 |
ba73744e5ae40e4eea8e6d4f1e2db5c36d1b983b03eb8fa883f4b9969bbcfe5e llama_webgpu_core.js
|
| 3 |
5be341eeb6fbf19f731fde155bb3afc71cf8553e0a8e6fc9ce10ad5917e792e5 llama_webgpu_core.wasm
|
| 4 |
47bbfa0fe897e708455b9497e9aa41d2e94489c329300c33b12936fb3169deca llama_webgpu_bridge_worker.js
|