NakliTechie commited on
Commit
79786d7
·
1 Parent(s): eb85f90

Sync from GitHub 2026-10-22T19:32:49Z

Browse files
Files changed (1) hide show
  1. index.html +28 -3
index.html CHANGED
@@ -2688,6 +2688,7 @@
2688
  <li><strong>Gemma 4 26B-A4B &middot; SSD-streamed (experimental)</strong> (~14 GB) &mdash; text, with reasoning. Google's Gemma 4 26B-A4B mixture-of-experts (128 experts, 8 active per token) from its official 4-bit QAT GGUF (14.4 GB) on the same from-scratch WebGPU engine family: attention, the dense MLP, routers and KV cache stay on the GPU (~1.6 GB plus a 4 GB expert pool) while the 3,840 routed experts stream from on-device storage (OPFS) as the router picks them. The first load copies the model into this browser's storage once (~14 GB of disk; about 12 minutes on a fast connection). About 23 tokens/s on an M4 Pro with 24 GB, and prompts are read in chunks at about 55 tokens/s. Greedy output matches llama.cpp. Flip <strong>Show reasoning</strong> for its thinking.</li>
2689
  <li><strong>LFM2.5 24B-A2B &middot; SSD-streamed (experimental)</strong> (~13.6 GB) &mdash; text, reasoning-only. Liquid AI's LFM2.5-24B-A2B (released 2026-10-13): a 24B mixture-of-experts (64 experts, 4 active per token, about 2.4B parameters per token) with short-convolution and attention layers, from Liquid's own 4-bit GGUF. Short-convolution and attention layers, routers and KV cache stay on the GPU while the 2,432 routed experts stream from on-device storage (OPFS) as the router picks them. The first load copies the model into this browser's storage once (~13.6 GB of disk). About 16 tokens/s on an M4 Pro with 24 GB, and prompts are read in chunks at about 50 tokens/s. Greedy output matches llama.cpp (8 of 8 prompts). It always reasons before it answers, so every reply opens with its reasoning. It can use tools (calculator, time, memory, reminders): it called the right tool on 6 of 6 test prompts and answered 5 of 5 plain questions without one.</li>
2690
  <li><strong>Flint 40B-A2.5B &middot; SSD-streamed (experimental)</strong> (~26 GB) &mdash; text, coding, reasoning. Recursive's Flint (released 2026-10-21, Apache-2.0): a 39.8B mixture-of-experts coding model (256 experts, 6 active per token, about 2.5B parameters per token), from Recursive's own 4-bit (Q4_1) GGUF, whose card asks for 48 GB of memory under llama.cpp. Attention, routers, the token table and KV cache stay on the GPU while the 12,032 routed experts stream from on-device storage (OPFS) as the router picks them. The first load copies the model into this browser's storage once (~26 GB of disk). About 8.7 tokens/s on an M4 Pro with 24 GB, and prompts are read at about 27 tokens/s. Its output stays as close to llama.cpp's Metal backend as llama.cpp's own CPU backend does. It reasons before it answers, so replies open with its reasoning. Tools are off for now.</li>
 
2691
  <li><strong>GLM-4.7-Flash 30B-A3B &middot; via anyformer (experimental)</strong> (~18 GB) &mdash; text, reasoning. Z.ai's GLM-4.7-Flash, a 30B mixture-of-experts with about 3B parameters per token, from unsloth's 4-bit GGUF (18.3 GB). LocalMind has no engine of its own for it: anyformer picks diskformer's llama.cpp path, which runs llama.cpp's WebGPU build in a worker and pages the experts in from on-device storage (OPFS). The first load copies the file into this browser's storage once. Flip <strong>Show reasoning</strong> for its thinking. Tools are off for now.</li>
2692
  <li><strong>Qwen3 4B &middot; FlashMap speculative (experimental)</strong> (~5.2 GB) &mdash; text, reasoning. Qwen's Qwen3-4B (Q8_0 GGUF, 4.3 GB) with FlashMap, a speculative-decoding drafter by Shen et al. (arXiv:2610.24365; 0.95 GB) that proposes 16 tokens in one step for the model to check in one pass. You get exactly the tokens the model would write on its own, sooner: on an M4 Pro about 2.1&ndash;2.2&times; on maths and 1.2&ndash;1.5&times; on code. Chat text drafts poorly, so there it falls back to plain decoding (about 0.7&ndash;1.1&times;). Both files are copied into this browser's storage once. Tools are off for now.</li>
2693
  </ul>
@@ -3006,6 +3007,7 @@
3006
  <option value="gemma4-26b-a4b-ssd">Gemma 4 26B-A4B &middot; SSD-streamed, experimental (~14 GB)</option>
3007
  <option value="lfm25-24b-a2b-ssd">LFM2.5 24B-A2B &middot; SSD-streamed, experimental (~13.6 GB)</option>
3008
  <option value="flint-40b-a2b-ssd">Flint 40B-A2.5B &middot; SSD-streamed, experimental, coding (~26 GB)</option>
 
3009
  <option value="glm-4.7-flash-anyformer">GLM-4.7-Flash 30B-A3B &middot; via anyformer, experimental (~18 GB)</option>
3010
  <option value="qwen3-4b-flashmap">Qwen3 4B &middot; FlashMap speculative decoding, experimental (~5.2 GB)</option>
3011
  </select>
@@ -9400,6 +9402,20 @@ self.addEventListener('message', async (e) => {
9400
  // backend:'anyformer-glm47flash' → loaded through anyformer (anyformer_models.js, jsDelivr tag v0.2.0), the first model
9401
  // LocalMind runs with no engine of its own (convergence plan, 2026-10-22). anyformer sends it to diskformer's
9402
  // llama.cpp path (llama.cpp WebGPU in a worker, experts paged from OPFS); unsloth's Q4_K_M GGUF, 18.3 GB.
 
 
 
 
 
 
 
 
 
 
 
 
 
 
9403
  'glm-4.7-flash-anyformer': {
9404
  id: 'unsloth/GLM-4.7-Flash-GGUF',
9405
  label: 'GLM-4.7-Flash 30B-A3B · via anyformer (experimental)',
@@ -17761,6 +17777,7 @@ self.addEventListener('message', async (e) => {
17761
  'gemma4moe-ssd': { engine: 'gemma4_moe_ssd.js', exportName: 'Gemma4MoeSsd', label: 'Gemma 4 26B-A4B (Q4_0)' },
17762
  'lfm2moe-ssd': { engine: 'lfm2_moe_ssd.js', exportName: 'Lfm2MoeSsd', label: 'LFM2.5 24B-A2B (Q4_0)' },
17763
  'flint-ssd': { engine: 'flint_moe_ssd.js', exportName: 'FlintMoeSsd', label: 'Flint 40B-A2.5B (Q4_1)' },
 
17764
  'anyformer-glm47flash': { engine: 'anyformer_models.js', exportName: 'Glm47Flash', label: 'GLM-4.7-Flash (Q4_K_M, via anyformer)' },
17765
  'qwen3dense': { engine: 'qwen35_dense.js', exportName: 'Qwen3Dense', label: 'Qwen3 4B (Q8_0) + FlashMap' },
17766
  'hyv3-ssd': { engine: 'hy_v3_moe_ssd.js', exportName: 'HyV3MoeSsd', label: 'Hy-Memory 30B-A3B (Q8_0)' }, // the memory writer only (WRITER_MODELS)
@@ -19378,7 +19395,7 @@ self.addEventListener('message', async (e) => {
19378
  const ENGINE_STORES = {
19379
  'transformers': ['transformers-cache'], 'wllama': ['wllama'], 'lfm2-webgpu': ['gguf-v1'],
19380
  'bonsai2-webgpu': ['gguf-cache-v1'], 'gemma4-webgpu': ['safetensors-cache-v1', 'bonsai-pipeline-v1'],
19381
- '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)'], 'qwen3dense': ['on disk (OPFS)'], 'qwen35dense': ['on disk (OPFS)'],
19382
  };
19383
  function parseSizeBytes(size) {
19384
  const m = /([\d.]+)\s*(GB|MB)/i.exec(String(size || ''));
@@ -24005,13 +24022,13 @@ self.addEventListener('message', async (e) => {
24005
  * { text }, reset(), dispose(), and the think markers. One subclass per model names its repo.
24006
  *
24007
  * anyformer loads from jsDelivr by GitHub tag until it is on npm. Through diskformer's llama.cpp path reasoning is off
24008
- * unless generate() passes { thinking: true } (anyformer v0.2.1+), and the reply is text only (no token ids). With
24009
  * reasoning on, the reply opens inside the thought; the class drops the literal <think> and </think> from the text
24010
  * and marks the boundary with a chunk whose token is THINK_CLOSE (the worker's close event).
24011
  */
24012
  const THINK_CLOSE = -2;
24013
 
24014
- export const ANYFORMER_URL = 'https://cdn.jsdelivr.net/gh/NakliTechie/anyformer.js@v0.2.2/index.js';
24015
 
24016
  // anyformer's progress events → the engine events the SSD worker posts: a download as 'weights', planning as 'init'.
24017
  function toEngineProgress(e) {
@@ -24067,6 +24084,14 @@ export class Glm47Flash extends AnyformerModel {
24067
  static repo = 'unsloth/GLM-4.7-Flash-GGUF';
24068
  static file = 'GLM-4.7-Flash-Q4_K_M.gguf';
24069
  }
 
 
 
 
 
 
 
 
24070
  </script>
24071
  <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
24072
  * message text and never on `instanceof` (two copies of this module, e.g. a CDN copy and a vendored one, make
 
2688
  <li><strong>Gemma 4 26B-A4B &middot; SSD-streamed (experimental)</strong> (~14 GB) &mdash; text, with reasoning. Google's Gemma 4 26B-A4B mixture-of-experts (128 experts, 8 active per token) from its official 4-bit QAT GGUF (14.4 GB) on the same from-scratch WebGPU engine family: attention, the dense MLP, routers and KV cache stay on the GPU (~1.6 GB plus a 4 GB expert pool) while the 3,840 routed experts stream from on-device storage (OPFS) as the router picks them. The first load copies the model into this browser's storage once (~14 GB of disk; about 12 minutes on a fast connection). About 23 tokens/s on an M4 Pro with 24 GB, and prompts are read in chunks at about 55 tokens/s. Greedy output matches llama.cpp. Flip <strong>Show reasoning</strong> for its thinking.</li>
2689
  <li><strong>LFM2.5 24B-A2B &middot; SSD-streamed (experimental)</strong> (~13.6 GB) &mdash; text, reasoning-only. Liquid AI's LFM2.5-24B-A2B (released 2026-10-13): a 24B mixture-of-experts (64 experts, 4 active per token, about 2.4B parameters per token) with short-convolution and attention layers, from Liquid's own 4-bit GGUF. Short-convolution and attention layers, routers and KV cache stay on the GPU while the 2,432 routed experts stream from on-device storage (OPFS) as the router picks them. The first load copies the model into this browser's storage once (~13.6 GB of disk). About 16 tokens/s on an M4 Pro with 24 GB, and prompts are read in chunks at about 50 tokens/s. Greedy output matches llama.cpp (8 of 8 prompts). It always reasons before it answers, so every reply opens with its reasoning. It can use tools (calculator, time, memory, reminders): it called the right tool on 6 of 6 test prompts and answered 5 of 5 plain questions without one.</li>
2690
  <li><strong>Flint 40B-A2.5B &middot; SSD-streamed (experimental)</strong> (~26 GB) &mdash; text, coding, reasoning. Recursive's Flint (released 2026-10-21, Apache-2.0): a 39.8B mixture-of-experts coding model (256 experts, 6 active per token, about 2.5B parameters per token), from Recursive's own 4-bit (Q4_1) GGUF, whose card asks for 48 GB of memory under llama.cpp. Attention, routers, the token table and KV cache stay on the GPU while the 12,032 routed experts stream from on-device storage (OPFS) as the router picks them. The first load copies the model into this browser's storage once (~26 GB of disk). About 8.7 tokens/s on an M4 Pro with 24 GB, and prompts are read at about 27 tokens/s. Its output stays as close to llama.cpp's Metal backend as llama.cpp's own CPU backend does. It reasons before it answers, so replies open with its reasoning. Tools are off for now.</li>
2691
+ <li><strong>Ling 3.0 Flash 124B-A5B &middot; via anyformer (experimental)</strong> (~81 GB) &mdash; text, reasoning. inclusionAI's Ling 3.0 Flash (released 2026-08-02, MIT): a 124B mixture-of-experts (512 experts, 8 active per token, about 5.1B parameters per token) whose layers mix linear attention (KDA) and latent attention (MLA), from inclusionAI's own 4-bit Q4_K_M GGUF (77 GB in two parts). anyformer loads it on diskformer's engine for it, which keeps the attention, routers and caches on the GPU (about 2.5 GB plus a 4 GB expert cache; it needs about 7 GB of GPU memory) and streams the 20,480 routed experts from on-device storage (OPFS). The first load copies the model into this browser's storage once (~81 GB of disk). About 5.5 tokens/s on an M4 Pro with 24 GB, and prompts are read at about 5 tokens/s. Flip <strong>Show reasoning</strong> for its thinking. Tools are off for now.</li>
2692
  <li><strong>GLM-4.7-Flash 30B-A3B &middot; via anyformer (experimental)</strong> (~18 GB) &mdash; text, reasoning. Z.ai's GLM-4.7-Flash, a 30B mixture-of-experts with about 3B parameters per token, from unsloth's 4-bit GGUF (18.3 GB). LocalMind has no engine of its own for it: anyformer picks diskformer's llama.cpp path, which runs llama.cpp's WebGPU build in a worker and pages the experts in from on-device storage (OPFS). The first load copies the file into this browser's storage once. Flip <strong>Show reasoning</strong> for its thinking. Tools are off for now.</li>
2693
  <li><strong>Qwen3 4B &middot; FlashMap speculative (experimental)</strong> (~5.2 GB) &mdash; text, reasoning. Qwen's Qwen3-4B (Q8_0 GGUF, 4.3 GB) with FlashMap, a speculative-decoding drafter by Shen et al. (arXiv:2610.24365; 0.95 GB) that proposes 16 tokens in one step for the model to check in one pass. You get exactly the tokens the model would write on its own, sooner: on an M4 Pro about 2.1&ndash;2.2&times; on maths and 1.2&ndash;1.5&times; on code. Chat text drafts poorly, so there it falls back to plain decoding (about 0.7&ndash;1.1&times;). Both files are copied into this browser's storage once. Tools are off for now.</li>
2694
  </ul>
 
3007
  <option value="gemma4-26b-a4b-ssd">Gemma 4 26B-A4B &middot; SSD-streamed, experimental (~14 GB)</option>
3008
  <option value="lfm25-24b-a2b-ssd">LFM2.5 24B-A2B &middot; SSD-streamed, experimental (~13.6 GB)</option>
3009
  <option value="flint-40b-a2b-ssd">Flint 40B-A2.5B &middot; SSD-streamed, experimental, coding (~26 GB)</option>
3010
+ <option value="ling-3-flash-anyformer">Ling 3.0 Flash 124B-A5B &middot; via anyformer, experimental (~81 GB)</option>
3011
  <option value="glm-4.7-flash-anyformer">GLM-4.7-Flash 30B-A3B &middot; via anyformer, experimental (~18 GB)</option>
3012
  <option value="qwen3-4b-flashmap">Qwen3 4B &middot; FlashMap speculative decoding, experimental (~5.2 GB)</option>
3013
  </select>
 
9402
  // backend:'anyformer-glm47flash' → loaded through anyformer (anyformer_models.js, jsDelivr tag v0.2.0), the first model
9403
  // LocalMind runs with no engine of its own (convergence plan, 2026-10-22). anyformer sends it to diskformer's
9404
  // llama.cpp path (llama.cpp WebGPU in a worker, experts paged from OPFS); unsloth's Q4_K_M GGUF, 18.3 GB.
9405
+ // backend:'anyformer-ling3flash' → Ling 3.0 Flash through anyformer, which sends it to diskformer's LingMoeSsd hand
9406
+ // engine (two-part Q4_K_M, 77.0 GB download, an 80.7 GB store under localmind-ssd/ling-3.0-flash-q4_k_m).
9407
+ 'ling-3-flash-anyformer': {
9408
+ id: 'inclusionAI/Ling-3.0-flash-GGUF',
9409
+ label: 'Ling 3.0 Flash 124B-A5B · via anyformer (experimental)',
9410
+ backend: 'anyformer-ling3flash',
9411
+ size: '~81 GB',
9412
+ type: 'causal',
9413
+ multimodal: false,
9414
+ agentCapable: false,
9415
+ thinking: true,
9416
+ contextSize: 4096,
9417
+ genConfig: { max_new_tokens: 2048 },
9418
+ },
9419
  'glm-4.7-flash-anyformer': {
9420
  id: 'unsloth/GLM-4.7-Flash-GGUF',
9421
  label: 'GLM-4.7-Flash 30B-A3B · via anyformer (experimental)',
 
17777
  'gemma4moe-ssd': { engine: 'gemma4_moe_ssd.js', exportName: 'Gemma4MoeSsd', label: 'Gemma 4 26B-A4B (Q4_0)' },
17778
  'lfm2moe-ssd': { engine: 'lfm2_moe_ssd.js', exportName: 'Lfm2MoeSsd', label: 'LFM2.5 24B-A2B (Q4_0)' },
17779
  'flint-ssd': { engine: 'flint_moe_ssd.js', exportName: 'FlintMoeSsd', label: 'Flint 40B-A2.5B (Q4_1)' },
17780
+ 'anyformer-ling3flash': { engine: 'anyformer_models.js', exportName: 'Ling3Flash', label: 'Ling 3.0 Flash (Q4_K_M, via anyformer)' },
17781
  'anyformer-glm47flash': { engine: 'anyformer_models.js', exportName: 'Glm47Flash', label: 'GLM-4.7-Flash (Q4_K_M, via anyformer)' },
17782
  'qwen3dense': { engine: 'qwen35_dense.js', exportName: 'Qwen3Dense', label: 'Qwen3 4B (Q8_0) + FlashMap' },
17783
  'hyv3-ssd': { engine: 'hy_v3_moe_ssd.js', exportName: 'HyV3MoeSsd', label: 'Hy-Memory 30B-A3B (Q8_0)' }, // the memory writer only (WRITER_MODELS)
 
19395
  const ENGINE_STORES = {
19396
  'transformers': ['transformers-cache'], 'wllama': ['wllama'], 'lfm2-webgpu': ['gguf-v1'],
19397
  'bonsai2-webgpu': ['gguf-cache-v1'], 'gemma4-webgpu': ['safetensors-cache-v1', 'bonsai-pipeline-v1'],
19398
+ '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)'], 'qwen3dense': ['on disk (OPFS)'], 'qwen35dense': ['on disk (OPFS)'],
19399
  };
19400
  function parseSizeBytes(size) {
19401
  const m = /([\d.]+)\s*(GB|MB)/i.exec(String(size || ''));
 
24022
  * { text }, reset(), dispose(), and the think markers. One subclass per model names its repo.
24023
  *
24024
  * anyformer loads from jsDelivr by GitHub tag until it is on npm. Through diskformer's llama.cpp path reasoning is off
24025
+ * unless generate() passes { thinking: true } (llama.cpp path since anyformer v0.2.1, hand engines since v0.2.4), and the reply is text only (no token ids). With
24026
  * reasoning on, the reply opens inside the thought; the class drops the literal <think> and </think> from the text
24027
  * and marks the boundary with a chunk whose token is THINK_CLOSE (the worker's close event).
24028
  */
24029
  const THINK_CLOSE = -2;
24030
 
24031
+ export const ANYFORMER_URL = 'https://cdn.jsdelivr.net/gh/NakliTechie/anyformer.js@v0.2.4/index.js';
24032
 
24033
  // anyformer's progress events → the engine events the SSD worker posts: a download as 'weights', planning as 'init'.
24034
  function toEngineProgress(e) {
 
24084
  static repo = 'unsloth/GLM-4.7-Flash-GGUF';
24085
  static file = 'GLM-4.7-Flash-Q4_K_M.gguf';
24086
  }
24087
+
24088
+ // inclusionAI's Ling 3.0 Flash (bailingmoe3, 124B-A5B): the two-part Q4_K_M; anyformer (v0.2.3+) sends it to diskformer's
24089
+ // LingMoeSsd hand engine (built in LocalMind 2026-10-22, now diskformer's; gate plan/2026-10-22-ling-harness), which
24090
+ // needs a GPU budget of 6.8 GB. With reasoning on, its template opens <think> in the prompt.
24091
+ export class Ling3Flash extends AnyformerModel {
24092
+ static repo = 'inclusionAI/Ling-3.0-flash-GGUF';
24093
+ static budget = 7e9;
24094
+ }
24095
  </script>
24096
  <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
24097
  * message text and never on `instanceof` (two copies of this module, e.g. a CDN copy and a vendored one, make