MLX
lora
tool-use
thoddnn commited on
Commit
e7f400a
·
verified ·
1 Parent(s): 700bf13

Publish verified incremental chat checkpoint 126 from production step 1900

Browse files
Files changed (3) hide show
  1. README.md +5 -1
  2. adapters.safetensors +1 -1
  3. manifest.json +6 -4
README.md CHANGED
@@ -11,6 +11,10 @@ tags:
11
 
12
  Qwen 3.5 2B coordinator adapter for Fenn’s shared chat SDK. This adapter generates bounded Python tool calls; it is not a general document-analysis adapter.
13
 
14
- Selected checkpoint: 1900. Rank 16; four adapted layers. Use the matching Nano base revision in manifest.json and Fenn’s current tool harness. Load the base without this adapter when using Nano as an analyst.
 
 
 
 
15
 
16
  Source base: [mlx-community/Qwen3.5-2B-8bit](https://huggingface.co/mlx-community/Qwen3.5-2B-8bit). The base retains its upstream Apache 2.0 license and attribution.
 
11
 
12
  Qwen 3.5 2B coordinator adapter for Fenn’s shared chat SDK. This adapter generates bounded Python tool calls; it is not a general document-analysis adapter.
13
 
14
+ Current release: incremental SFT checkpoint 126, continued from production checkpoint 1900. Rank 16; four adapted layers; 8-bit base. This release adds indexing and duplicate-scope demonstrations with replay examples.
15
+
16
+ Use the matching Nano base revision in `manifest.json` and Fenn’s current tool harness. Load the base without this adapter when using Nano as an analyst. Manifest checksums cover the adapter configuration and weights.
17
+
18
+ The adapter passed integrity and fresh-process reload checks. A five-case focused tool-use evaluation improved from one to four passing cases; this small evaluation is not a general quality benchmark. Duplicate scanning without an explicit folder can still produce an internal workspace path, which the current Fenn harness redirects to known indexed folders.
19
 
20
  Source base: [mlx-community/Qwen3.5-2B-8bit](https://huggingface.co/mlx-community/Qwen3.5-2B-8bit). The base retains its upstream Apache 2.0 license and attribution.
adapters.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:7394bb3f6a9f8674b98559f4f6d244fb6ea65bc5ccf2c748fcf239f954989df8
3
  size 11220564
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b1ddba57b493965c5e05ef7d6808fb31e6d48fe6d385a931f3f5cad8cacceda1
3
  size 11220564
manifest.json CHANGED
@@ -1,12 +1,12 @@
1
  {
2
  "schema": "fenn-chat-adapter-v1",
3
- "id": "terra-fresh-step-1900",
4
  "base_repo": "thoddnn/Qwen3.5-2B-8bit",
5
  "base_revision": "dc098b420dc52743e621579391d40ef26e2b9d06",
6
  "source_repo": "mlx-community/Qwen3.5-2B-8bit",
7
  "source_revision": "562fb8ea19bbe7565a3ddf5b7d49899ea2c88d2b",
8
  "quantization_bits": 8,
9
- "checkpoint": 1900,
10
  "files": {
11
  "adapter_config.json": {
12
  "bytes": 132,
@@ -14,7 +14,9 @@
14
  },
15
  "adapters.safetensors": {
16
  "bytes": 11220564,
17
- "sha256": "7394bb3f6a9f8674b98559f4f6d244fb6ea65bc5ccf2c748fcf239f954989df8"
18
  }
19
- }
 
 
20
  }
 
1
  {
2
  "schema": "fenn-chat-adapter-v1",
3
+ "id": "production-scope-incremental-2026-11-14",
4
  "base_repo": "thoddnn/Qwen3.5-2B-8bit",
5
  "base_revision": "dc098b420dc52743e621579391d40ef26e2b9d06",
6
  "source_repo": "mlx-community/Qwen3.5-2B-8bit",
7
  "source_revision": "562fb8ea19bbe7565a3ddf5b7d49899ea2c88d2b",
8
  "quantization_bits": 8,
9
+ "checkpoint": 126,
10
  "files": {
11
  "adapter_config.json": {
12
  "bytes": 132,
 
14
  },
15
  "adapters.safetensors": {
16
  "bytes": 11220564,
17
+ "sha256": "b1ddba57b493965c5e05ef7d6808fb31e6d48fe6d385a931f3f5cad8cacceda1"
18
  }
19
+ },
20
+ "parent_checkpoint": 1900,
21
+ "parent_adapter_sha256": "7394bb3f6a9f8674b98559f4f6d244fb6ea65bc5ccf2c748fcf239f954989df8"
22
  }