README: mask_locality layout entry, q3_8b license row
Browse files
README.md
CHANGED
|
@@ -14,6 +14,7 @@ license_name: per-model
|
|
| 14 |
- `<model>/manifest.json`: file list, sha256 of every file and tensor, restore paths
|
| 15 |
- `<model>/ppl/`: tokenized WikiText-2 / C4 evaluation windows (int32 token ids) with their `manifest.json`
|
| 16 |
- `<model>/moments/`: calibration second-moment statistics the key-side bases were fitted from
|
|
|
|
| 17 |
- **Format (.lqz)**: a 4-bit group of 128 values holds at most 16 distinct fp16 values, stored as a 16-entry fp16 table plus
|
| 18 |
128 4-bit indices; all other tensors are stored raw. Decoding is exact (bit-identical tensors).
|
| 19 |
- **Not** an inference-ready release: research artifacts only, no quality guarantee.
|
|
@@ -25,6 +26,7 @@ Each folder carries the license of the model it is derived from.
|
|
| 25 |
| Folder | Derived from | License |
|
| 26 |
|---|---|---|
|
| 27 |
| `q3_4b/` | `Qwen/Qwen3-4B` | Apache-2.0 |
|
|
|
|
| 28 |
| `l31i/` | `meta-llama/Llama-3.1-8B-Instruct` | Llama 3.1 Community License (`l31i/LICENSE`, `l31i/USE_POLICY.md`) |
|
| 29 |
| `l31/` | `meta-llama/Llama-3.1-8B` | Llama 3.1 Community License (`l31/LICENSE`, `l31/USE_POLICY.md`) |
|
| 30 |
| `l2/` | `meta-llama/Llama-2-7b-hf` | Llama 2 Community License (`l2/LICENSE.txt`, `l2/USE_POLICY.md`) |
|
|
|
|
| 14 |
- `<model>/manifest.json`: file list, sha256 of every file and tensor, restore paths
|
| 15 |
- `<model>/ppl/`: tokenized WikiText-2 / C4 evaluation windows (int32 token ids) with their `manifest.json`
|
| 16 |
- `<model>/moments/`: calibration second-moment statistics the key-side bases were fitted from
|
| 17 |
+
- `<model>/mask_locality/cache/`: inputs and intermediate products of a pruning-mask locality study (channel permutations from calibration, per-channel keep-frequency profiles); sha256 of every file is recorded in the study's `ARTIFACTS.json`
|
| 18 |
- **Format (.lqz)**: a 4-bit group of 128 values holds at most 16 distinct fp16 values, stored as a 16-entry fp16 table plus
|
| 19 |
128 4-bit indices; all other tensors are stored raw. Decoding is exact (bit-identical tensors).
|
| 20 |
- **Not** an inference-ready release: research artifacts only, no quality guarantee.
|
|
|
|
| 26 |
| Folder | Derived from | License |
|
| 27 |
|---|---|---|
|
| 28 |
| `q3_4b/` | `Qwen/Qwen3-4B` | Apache-2.0 |
|
| 29 |
+
| `q3_8b/` | `Qwen/Qwen3-8B` | Apache-2.0 |
|
| 30 |
| `l31i/` | `meta-llama/Llama-3.1-8B-Instruct` | Llama 3.1 Community License (`l31i/LICENSE`, `l31i/USE_POLICY.md`) |
|
| 31 |
| `l31/` | `meta-llama/Llama-3.1-8B` | Llama 3.1 Community License (`l31/LICENSE`, `l31/USE_POLICY.md`) |
|
| 32 |
| `l2/` | `meta-llama/Llama-2-7b-hf` | Llama 2 Community License (`l2/LICENSE.txt`, `l2/USE_POLICY.md`) |
|