Update README.md
Browse files
README.md
CHANGED
|
@@ -1,6 +1,12 @@
|
|
| 1 |
---
|
| 2 |
base_model:
|
| 3 |
- qingy2024/Qwarkstar-4B
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 4 |
---
|
| 5 |
|
| 6 |
## Qwarkstar 4B Instruct (Preview)
|
|
@@ -8,6 +14,11 @@ base_model:
|
|
| 8 |
> [!NOTE]
|
| 9 |
> Training complete!
|
| 10 |
|
| 11 |
-
|
|
|
|
| 12 |
|
| 13 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
---
|
| 2 |
base_model:
|
| 3 |
- qingy2024/Qwarkstar-4B
|
| 4 |
+
license: apache-2.0
|
| 5 |
+
datasets:
|
| 6 |
+
- HuggingFaceTB/smoltalk
|
| 7 |
+
language:
|
| 8 |
+
- en
|
| 9 |
+
pipeline_tag: text-generation
|
| 10 |
---
|
| 11 |
|
| 12 |
## Qwarkstar 4B Instruct (Preview)
|
|
|
|
| 14 |
> [!NOTE]
|
| 15 |
> Training complete!
|
| 16 |
|
| 17 |
+
This model is fine-tuned using Supervised Fine-Tuning (SFT) on 100k samples from the `HuggingFaceTB/smoltalk` dataset.
|
| 18 |
+
It follows the ChatML input-output formatting template.
|
| 19 |
|
| 20 |
+
### Training Details:
|
| 21 |
+
- **Base Model**: `qingy2024/Qwarkstar-4B`
|
| 22 |
+
- **Batch Size**: 32 (2 H100s x 8 per GPU)
|
| 23 |
+
- **Max Gradient Norm**: 1.0
|
| 24 |
+
- **Final Loss**: ~0.59
|