anyformer 0.4.16 → 0.4.17

package/CHANGELOG.md

@@ -1,5 +1,19 @@
 # Changelog
 
+## 0.4.17 — 2026-10-30
+
+### Added
+- `MODELS.chat.CLOVER_35B_A3B`: Diffuspace's Clover-35B-A3B-Preview (Apache-2.0), block diffusion on Qwen3.6-35B-A3B,
+  text only, on diskformer's `clover_moe_ssd.js` (Q4_0 GGUF, 19.6 GB, from OPFS; 4 GB budget). `diffusion` and
+  `experimental`; speed unmeasured. generate() yields once per committed 32-token block. Hub fixture at revision
+  32e22a0. Browser test `chat` passed (Chrome, this Mac, 23:17 IST, GGUF from Models2, 4 GB budget): load 55.6 s
+  including the copy into OPFS, an on-topic 25-token reply in 25.4 s.
+- Browser test `chat` takes `?local=1` for a diskformer GGUF served from `--models`.
+
+### Changed
+- diskformer pin 93193d0 → f17009c (v0.6.15: the Clover engine; `qwen35_moe_ssd.js` and `qwen3_moe_ssd.js` only learn
+  to accept the `clover_moe` architecture).
+
 ## 0.4.16 — 2026-10-30
 
 ### Changed

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "anyformer",
-  "version": "0.4.16",
+  "version": "0.4.17",
   "description": "One call to run a model in a browser tab: anyformer reads the Hub repo, probes the browser, and picks the back end (transformers.js, LocalMind's WebGPU kernels, diskformer, kohra) that fits the model's format, architecture and your GPU budget.",
   "type": "module",
   "main": "./index.js",

package/src/backends/diskformer.js

@@ -27,6 +27,17 @@ export const DISKFORMER_MODELS = [
     engine: 'qwen35_moe_ssd.js', Engine: 'MiniCpmV47Ssd', minBudget: 4e9,
     speeds: [[6.1e9, 14.1]],
   },
+  // Clover-35B-A3B-Preview (Diffuspace, Apache-2.0): block diffusion on Qwen3.6-35B-A3B, text only. The engine
+  // (engines/clover_moe_ssd.js CLOVER_35B_A3B, rev 32e22a0) fills a 32-token block over several denoising passes
+  // (kohra's denoise module), so generate() yields once per committed block. Gated by diskformer (v0.6.15: oracle
+  // and browser chat) at a 1 GB expert pool and maxCtx 1024; speed unmeasured. At maxCtx 1024 the engine fixes about
+  // 2.0 GB (dense 1.46, KV and DeltaNet state 0.09, working buffers) and refuses under about 2.05 GB (diskformer's
+  // reading of its budget formula); 4 GB, Qwen3.6's floor, leaves about 1.9 GB of experts.
+  {
+    repo: 'Diffuspace/Clover-35B-A3B-Preview-GGUF', file: 'Clover-35B-A3B-Preview-Q4_0.gguf', size: 19586909600,
+    engine: 'clover_moe_ssd.js', Engine: 'CloverMoeSsd', minBudget: 4e9,
+    speeds: [],
+  },
   // From LocalMind's gates (M4 Pro, 4 GiB pool; plan/2026-10-22-flint-harness): Flint ~7.3 GB on the GPU at maxCtx
   // 8192 (dense 2.33, pool 4.29, KV 0.41, batch 0.20), 8.66 tok/s at lookahead 1. LFM2.5 24B ~16 tok/s at a 4 GiB
   // pool; its total GPU use was not recorded, so it carries no speed estimate. The smallest working budget of either

package/src/models.js

@@ -1,7 +1,7 @@
 // Models that pass a browser test through anyformer, grouped by what they do. Pass an entry to load() in place of a
 // repo id; it carries the task and, where the back end needs one, the budget. DEFAULT is the smallest download that
 // passed. `test` names the page under test/browser/ that proves the entry; test/models.test.js checks it exists.
-// `diffusion: true` marks a text-diffusion model (the reply denoises in blocks); `experimental: true` marks an entry its
+// `diffusion: true` marks a text-diffusion model (the reply denoises in blocks, so generate() chunks arrive a block at a time); `experimental: true` marks an entry its
 // app shows as opt-in until that app's own checks pass; `audioOut: true` marks a model that can also answer in speech
 // (generate({ speech: true })).
 
@@ -21,6 +21,8 @@ const MINICPM5_2B = M('openbmb/MiniCPM5-2B-GGUF', 'wllama', 'text-generation', '
 const QWEN3_0_6B_MDLM = M('naklitechie/Qwen3-0.6B-diffusion-mdlm-ONNX', 'kohra', 'text-generation', 'kohra', { downloadBytes: 1.54e9, diffusion: true });
 const NEMOTRON_DIFFUSION_3B = M('naklitechie/Nemotron-Labs-Diffusion-3B-ONNX', 'kohra', 'text-generation', 'kohra', { downloadBytes: 2.77e9, diffusion: true, experimental: true });
 const GEMMA4_26B_A4B = M('google/gemma-4-26B-A4B-it-qat-q4_0-gguf', 'diskformer', 'text-generation', 'diskformer-chat', { budget: 2.5e9, downloadBytes: 14.44e9 });
+// Block diffusion on Qwen3.6-35B-A3B: replies arrive one 32-token block at a time. Speed unmeasured.
+const CLOVER_35B_A3B = M('Diffuspace/Clover-35B-A3B-Preview-GGUF', 'diskformer', 'text-generation', 'chat', { budget: 4e9, downloadBytes: 19.59e9, diffusion: true, experimental: true });
 const LFM3_1_7B = M('LiquidAI/LFM3-1.7B-ONNX', 'transformers', 'text-generation', 'chat', { downloadBytes: 1.63e9 });
 // transformers.js 4.3.2 has no lfm3_omni class, so diskformer's hand engine runs it; swap the entry to transformers.js
 // when it adds the model.
@@ -57,6 +59,8 @@ export const MODELS = {
     NEMOTRON_DIFFUSION_3B,
     /** 26B parameters in 2.5 GB of GPU memory; 14.4 GB on disk (diskformer). */
     GEMMA4_26B_A4B,
+    /** Diffuspace's Clover-35B-A3B-Preview: block diffusion on Qwen3.6-35B-A3B, 19.6 GB on disk; replies arrive in 32-token blocks (diskformer). */
+    CLOVER_35B_A3B,
     /** NVIDIA's Nemotron 3.5 Bolt E4B (dense, text only), a community q4f16 ONNX export in Gemma 4 E4B's graph. */
     NEMOTRON_3_5_BOLT_E4B,
     /** LiquidAI's LFM3 1.7B (dense), LiquidAI's own ONNX export; thinking off unless a request asks for it. */

package/src/sources.js

@@ -9,6 +9,8 @@ export const PINS = {
   // v0.3.1: seeded block-KV graphs (Nemotron-Labs-Diffusion-3B), chatFormat, split external data; download progress
   // (progress_callback) and the graph files cached in Cache Storage 'kohra-models'.
   kohra: 'bf95d44c3e3cfc001201baa03fa121c2bf9c7d73',
+  // v0.6.15 (f17009c): engines/clover_moe_ssd.js, Clover-35B-A3B-Preview (block diffusion on Qwen3.6-35B-A3B; kohra's
+  // denoise loop vendored in engines/vendor/). qwen35_moe_ssd.js and qwen3_moe_ssd.js only learn to accept clover_moe.
   // v0.6.14 (93193d0): LFM3-Omni speech out's inverse STFT uses an FFT (same samples), 0.54x → 1.5–1.8x real time on
   // an M4 Pro. Changes only engines/lfm3_omni.js.
   // v0.6.13 (7ffbea5): LFM3-Omni serves one reply at a time; overlapping runs broke onnxruntime-web WebGPU.
@@ -21,7 +23,7 @@ export const PINS = {
   // Gemma 4 E2B/E4B and Ternary Bonsai 2) under engines/, and the llama.cpp path under ggml/, which loads split GGUFs
   // and drafters, and reads past 2 GiB (Nemotron 3.5 Bolt E4B needs v0.6.8 or later). The hand engines' ingest
   // streams the GGUF header with progress and resumes a dropped range.
-  diskformer: '93193d0913f5e76059fa45db509dc670e7527430',
+  diskformer: 'f17009cc31e15361af7ae5536e8f55cb43c5066f',
 };
 
 const gh = (repo, sha, path) => `https://cdn.jsdelivr.net/gh/NakliTechie/${repo}@${sha}/${path}`;