Update README.md
Browse files
README.md
CHANGED
|
@@ -1,13 +1,28 @@
|
|
| 1 |
---
|
| 2 |
-
language: en
|
| 3 |
-
library_name: mlx
|
| 4 |
pipeline_tag: text-generation
|
|
|
|
|
|
|
|
|
|
|
|
|
| 5 |
tags:
|
| 6 |
- mlx
|
|
|
|
| 7 |
---
|
| 8 |
|
| 9 |
# catalystsec/MiniMax-M2.7-4bit-DWQ
|
| 10 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 11 |
## Use with mlx
|
| 12 |
|
| 13 |
```bash
|
|
|
|
| 1 |
---
|
|
|
|
|
|
|
| 2 |
pipeline_tag: text-generation
|
| 3 |
+
license: other
|
| 4 |
+
license_name: modified-mit
|
| 5 |
+
license_link: https://github.com/MiniMax-AI/MiniMax-M2.7/blob/main/LICENSE
|
| 6 |
+
library_name: mlx
|
| 7 |
tags:
|
| 8 |
- mlx
|
| 9 |
+
base_model: MiniMaxAI/MiniMax-M2.7
|
| 10 |
---
|
| 11 |
|
| 12 |
# catalystsec/MiniMax-M2.7-4bit-DWQ
|
| 13 |
|
| 14 |
+
This model was quantized to 4-bit using DWQ with mlx-lm version **0.31.2**.
|
| 15 |
+
|
| 16 |
+
| Parameter | Value |
|
| 17 |
+
|---------------------------|--------------------------------|
|
| 18 |
+
| DWQ learning rate | 2e-7 |
|
| 19 |
+
| Batch size | 1 |
|
| 20 |
+
| Dataset | `allenai/tulu-3-sft-mixture` |
|
| 21 |
+
| Initial validation loss | 0.051 |
|
| 22 |
+
| Final validation loss | 0.032 |
|
| 23 |
+
| Relative KL reduction | ≈37 % |
|
| 24 |
+
| Tokens processed | ≈1.16 M |
|
| 25 |
+
|
| 26 |
## Use with mlx
|
| 27 |
|
| 28 |
```bash
|