Spaces:
Running
Running
NakliTechie commited on
Commit ·
2b9f2e1
1
Parent(s): 8d9f687
Sync from GitHub 2026-10-22T22:56:58Z
Browse files- index.html +14 -2
index.html
CHANGED
|
@@ -9351,10 +9351,12 @@ self.addEventListener('message', async (e) => {
|
|
| 9351 |
// MLP, routers and KV stay on the GPU (~1.6 GB + a 4 GB expert pool); the 3,840 routed experts (3.3 MB
|
| 9352 |
// each) stream from OPFS. Experimental: rung 2c of the SSD-streaming ladder, ~23 tok/s decode and ~55 tok/s
|
| 9353 |
// chunked prefill on an M4 Pro. Greedy output matches llama.cpp Metal 8/8 on the full model.
|
|
|
|
|
|
|
| 9354 |
'gemma4-26b-a4b-ssd': {
|
| 9355 |
id: 'google/gemma-4-26B-A4B-it-qat-q4_0-gguf',
|
| 9356 |
label: 'Gemma 4 26B-A4B · SSD-streamed (experimental)',
|
| 9357 |
-
backend: '
|
| 9358 |
size: '~14 GB',
|
| 9359 |
gguf: { file: 'gemma-4-26B_q4_0-it.gguf', size: 14439363584 }, // GEMMA4_26B_A4B in gemma4_moe_ssd.js
|
| 9360 |
type: 'causal',
|
|
@@ -17779,6 +17781,7 @@ self.addEventListener('message', async (e) => {
|
|
| 17779 |
'gemma4moe-ssd': { engine: 'gemma4_moe_ssd.js', exportName: 'Gemma4MoeSsd', label: 'Gemma 4 26B-A4B (Q4_0)' },
|
| 17780 |
'lfm2moe-ssd': { engine: 'lfm2_moe_ssd.js', exportName: 'Lfm2MoeSsd', label: 'LFM2.5 24B-A2B (Q4_0)' },
|
| 17781 |
'flint-ssd': { engine: 'flint_moe_ssd.js', exportName: 'FlintMoeSsd', label: 'Flint 40B-A2.5B (Q4_1)' },
|
|
|
|
| 17782 |
'anyformer-qwen36': { engine: 'anyformer_models.js', exportName: 'Qwen36', label: 'Qwen3.6 35B-A3B (Q8_0, via anyformer)' },
|
| 17783 |
'anyformer-ling3flash': { engine: 'anyformer_models.js', exportName: 'Ling3Flash', label: 'Ling 3.0 Flash (Q4_K_M, via anyformer)' },
|
| 17784 |
'anyformer-glm47flash': { engine: 'anyformer_models.js', exportName: 'Glm47Flash', label: 'GLM-4.7-Flash (Q4_K_M, via anyformer)' },
|
|
@@ -19398,7 +19401,7 @@ self.addEventListener('message', async (e) => {
|
|
| 19398 |
const ENGINE_STORES = {
|
| 19399 |
'transformers': ['transformers-cache'], 'wllama': ['wllama'], 'lfm2-webgpu': ['gguf-v1'],
|
| 19400 |
'bonsai2-webgpu': ['gguf-cache-v1'], 'gemma4-webgpu': ['safetensors-cache-v1', 'bonsai-pipeline-v1'],
|
| 19401 |
-
'qwen35moe-ssd': ['on disk (OPFS)'], 'minicpmv-ssd': ['on disk (OPFS)'], 'gemma4moe-ssd': ['on disk (OPFS)'], 'lfm2moe-ssd': ['on disk (OPFS)'], 'flint-ssd': ['on disk (OPFS)'], 'anyformer-glm47flash': ['on disk (OPFS)'], 'anyformer-ling3flash': ['on disk (OPFS)'], 'anyformer-qwen36': ['on disk (OPFS)'], 'qwen3dense': ['on disk (OPFS)'], 'qwen35dense': ['on disk (OPFS)'],
|
| 19402 |
};
|
| 19403 |
function parseSizeBytes(size) {
|
| 19404 |
const m = /([\d.]+)\s*(GB|MB)/i.exec(String(size || ''));
|
|
@@ -24124,6 +24127,15 @@ export class Qwen36 extends AnyformerModel {
|
|
| 24124 |
static budget = 9e9;
|
| 24125 |
static poolBytes = 4 * 2 ** 30;
|
| 24126 |
}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 24127 |
</script>
|
| 24128 |
<script type="text/plain" data-module="errors.js">/* errors.js — coded errors for the store, ingest and the engines. A host branches on `code`, never on the
|
| 24129 |
* message text and never on `instanceof` (two copies of this module, e.g. a CDN copy and a vendored one, make
|
|
|
|
| 9351 |
// MLP, routers and KV stay on the GPU (~1.6 GB + a 4 GB expert pool); the 3,840 routed experts (3.3 MB
|
| 9352 |
// each) stream from OPFS. Experimental: rung 2c of the SSD-streaming ladder, ~23 tok/s decode and ~55 tok/s
|
| 9353 |
// chunked prefill on an M4 Pro. Greedy output matches llama.cpp Metal 8/8 on the full model.
|
| 9354 |
+
// Since 2026-10-23 loaded through anyformer (backend 'anyformer-gemma26b': Gemma4_26B in anyformer_models.js),
|
| 9355 |
+
// which runs diskformer's copy of the same engine; the store under localmind-ssd/ is reused.
|
| 9356 |
'gemma4-26b-a4b-ssd': {
|
| 9357 |
id: 'google/gemma-4-26B-A4B-it-qat-q4_0-gguf',
|
| 9358 |
label: 'Gemma 4 26B-A4B · SSD-streamed (experimental)',
|
| 9359 |
+
backend: 'anyformer-gemma26b',
|
| 9360 |
size: '~14 GB',
|
| 9361 |
gguf: { file: 'gemma-4-26B_q4_0-it.gguf', size: 14439363584 }, // GEMMA4_26B_A4B in gemma4_moe_ssd.js
|
| 9362 |
type: 'causal',
|
|
|
|
| 17781 |
'gemma4moe-ssd': { engine: 'gemma4_moe_ssd.js', exportName: 'Gemma4MoeSsd', label: 'Gemma 4 26B-A4B (Q4_0)' },
|
| 17782 |
'lfm2moe-ssd': { engine: 'lfm2_moe_ssd.js', exportName: 'Lfm2MoeSsd', label: 'LFM2.5 24B-A2B (Q4_0)' },
|
| 17783 |
'flint-ssd': { engine: 'flint_moe_ssd.js', exportName: 'FlintMoeSsd', label: 'Flint 40B-A2.5B (Q4_1)' },
|
| 17784 |
+
'anyformer-gemma26b': { engine: 'anyformer_models.js', exportName: 'Gemma4_26B', label: 'Gemma 4 26B-A4B (Q4_0, via anyformer)' },
|
| 17785 |
'anyformer-qwen36': { engine: 'anyformer_models.js', exportName: 'Qwen36', label: 'Qwen3.6 35B-A3B (Q8_0, via anyformer)' },
|
| 17786 |
'anyformer-ling3flash': { engine: 'anyformer_models.js', exportName: 'Ling3Flash', label: 'Ling 3.0 Flash (Q4_K_M, via anyformer)' },
|
| 17787 |
'anyformer-glm47flash': { engine: 'anyformer_models.js', exportName: 'Glm47Flash', label: 'GLM-4.7-Flash (Q4_K_M, via anyformer)' },
|
|
|
|
| 19401 |
const ENGINE_STORES = {
|
| 19402 |
'transformers': ['transformers-cache'], 'wllama': ['wllama'], 'lfm2-webgpu': ['gguf-v1'],
|
| 19403 |
'bonsai2-webgpu': ['gguf-cache-v1'], 'gemma4-webgpu': ['safetensors-cache-v1', 'bonsai-pipeline-v1'],
|
| 19404 |
+
'qwen35moe-ssd': ['on disk (OPFS)'], 'minicpmv-ssd': ['on disk (OPFS)'], 'gemma4moe-ssd': ['on disk (OPFS)'], 'lfm2moe-ssd': ['on disk (OPFS)'], 'flint-ssd': ['on disk (OPFS)'], 'anyformer-glm47flash': ['on disk (OPFS)'], 'anyformer-ling3flash': ['on disk (OPFS)'], 'anyformer-qwen36': ['on disk (OPFS)'], 'anyformer-gemma26b': ['on disk (OPFS)'], 'qwen3dense': ['on disk (OPFS)'], 'qwen35dense': ['on disk (OPFS)'],
|
| 19405 |
};
|
| 19406 |
function parseSizeBytes(size) {
|
| 19407 |
const m = /([\d.]+)\s*(GB|MB)/i.exec(String(size || ''));
|
|
|
|
| 24127 |
static budget = 9e9;
|
| 24128 |
static poolBytes = 4 * 2 ** 30;
|
| 24129 |
}
|
| 24130 |
+
|
| 24131 |
+
// Gemma 4 26B-A4B: Google's QAT Q4_0 GGUF (14.4 GB) on diskformer's Gemma4MoeSsd hand engine, which LocalMind ran
|
| 24132 |
+
// itself until 2026-10-23; the stored copy is reused. Its thought channel (<|channel>thought … <channel|>) comes
|
| 24133 |
+
// through session.think (preamble 'thought\n'). A 4 GiB expert cache as before.
|
| 24134 |
+
export class Gemma4_26B extends AnyformerModel {
|
| 24135 |
+
static repo = 'google/gemma-4-26B-A4B-it-qat-q4_0-gguf';
|
| 24136 |
+
static budget = 7.5e9;
|
| 24137 |
+
static poolBytes = 4 * 2 ** 30;
|
| 24138 |
+
}
|
| 24139 |
</script>
|
| 24140 |
<script type="text/plain" data-module="errors.js">/* errors.js — coded errors for the store, ingest and the engines. A host branches on `code`, never on the
|
| 24141 |
* message text and never on `instanceof` (two copies of this module, e.g. a CDN copy and a vendored one, make
|