anyformer 0.4.18 → 0.4.19
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/CHANGELOG.md +16 -0
- package/package.json +1 -1
- package/src/sources.js +3 -1
package/CHANGELOG.md
CHANGED
|
@@ -1,5 +1,21 @@
|
|
|
1
1
|
# Changelog
|
|
2
2
|
|
|
3
|
+
## 0.4.19 — 2026-10-31
|
|
4
|
+
|
|
5
|
+
### Changed
|
|
6
|
+
- diskformer pin 314362e → 317d2a0 (v0.6.18): with thinking off, diskformer's llama.cpp chat closes a chat template's
|
|
7
|
+
always-open `<think>` block and bans the `<think>` token for the reply, so LFM2.5-24B-A2B (base and
|
|
8
|
+
`LFM2_5_24B_PI`) answers directly. This resolves 0.4.18's known limitation. diskformer's gates (Chrome, 4 GB):
|
|
9
|
+
thinking off 9/9 plain replies with and without the Pi adapter; thinking on unchanged (8/9 against llama.cpp).
|
|
10
|
+
|
|
11
|
+
### Verified
|
|
12
|
+
- `LFM2_5_24B_PI` through anyformer, 4 turns, thinking off (browser `chat?entry=chat.LFM2_5_24B_PI&local=1&turns=4&tokens=96`,
|
|
13
|
+
Chrome 156, this Mac, 12:12 IST, base GGUF and adapter from local files, diskformer 317d2a0): PASS. Loaded in 101.1 s;
|
|
14
|
+
every reply on topic with no `<think>` text (CPU, RAM, which is faster with latency figures, a one-sentence
|
|
15
|
+
summary); 2.6–3.1 tok/s.
|
|
16
|
+
- LFM2.5 230M, 4 turns, on 0.4.18's pin 314362e (browser `chat?local=1&turns=4`, `localmind`, 12:13 IST): PASS, load
|
|
17
|
+
8.7 s, 4/4 turns on topic.
|
|
18
|
+
|
|
3
19
|
## 0.4.18 — 2026-10-31
|
|
4
20
|
|
|
5
21
|
### Added
|
package/package.json
CHANGED
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
{
|
|
2
2
|
"name": "anyformer",
|
|
3
|
-
"version": "0.4.
|
|
3
|
+
"version": "0.4.19",
|
|
4
4
|
"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.",
|
|
5
5
|
"type": "module",
|
|
6
6
|
"main": "./index.js",
|
package/src/sources.js
CHANGED
|
@@ -9,6 +9,8 @@ export const PINS = {
|
|
|
9
9
|
// v0.3.1: seeded block-KV graphs (Nemotron-Labs-Diffusion-3B), chatFormat, split external data; download progress
|
|
10
10
|
// (progress_callback) and the graph files cached in Cache Storage 'kohra-models'.
|
|
11
11
|
kohra: 'bf95d44c3e3cfc001201baa03fa121c2bf9c7d73',
|
|
12
|
+
// v0.6.18 (317d2a0): with thinking off, the llama.cpp chat closes a template's always-open <think> (LFM2.5-24B) and
|
|
13
|
+
// bans the <think> token for the reply.
|
|
12
14
|
// v0.6.17 (314362e): LoRA adapters on the llama.cpp path (llama.load({ lora, loraScale })), and lfm2_moe_ssd.js runs
|
|
13
15
|
// a LoRA-merged LFM2.5-24B-A2B with Q8_0 dense matrices.
|
|
14
16
|
// v0.6.16 (467052f): engines/subgroup-matrix-guard.js. Chrome 156 rejects mixed f16/f32 subgroup-matrix kernels
|
|
@@ -28,7 +30,7 @@ export const PINS = {
|
|
|
28
30
|
// Gemma 4 E2B/E4B and Ternary Bonsai 2) under engines/, and the llama.cpp path under ggml/, which loads split GGUFs
|
|
29
31
|
// and drafters, and reads past 2 GiB (Nemotron 3.5 Bolt E4B needs v0.6.8 or later). The hand engines' ingest
|
|
30
32
|
// streams the GGUF header with progress and resumes a dropped range.
|
|
31
|
-
diskformer: '
|
|
33
|
+
diskformer: '317d2a002d95278e504ed84fab2d0f71dabaa884',
|
|
32
34
|
};
|
|
33
35
|
|
|
34
36
|
const gh = (repo, sha, path) => `https://cdn.jsdelivr.net/gh/NakliTechie/${repo}@${sha}/${path}`;
|