Commit
•
4648012
1
Parent(s):
1dbbf47
Update README.md
Browse files
README.md
CHANGED
|
@@ -13,4 +13,12 @@ Instruction tuned with the following parameters:
|
|
| 13 |
- LORA, Rank 8, Alpha 16, Dropout 0.05, all modules (QKV and MLP)
|
| 14 |
- 3 epochs
|
| 15 |
- Micro Batch Size 32 over 4xH100, gradient accumulation steps = 1
|
| 16 |
-
- AdamW with learning rate 5e-5
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 13 |
- LORA, Rank 8, Alpha 16, Dropout 0.05, all modules (QKV and MLP)
|
| 14 |
- 3 epochs
|
| 15 |
- Micro Batch Size 32 over 4xH100, gradient accumulation steps = 1
|
| 16 |
+
- AdamW with learning rate 5e-5
|
| 17 |
+
|
| 18 |
+
### Evaluation Results
|
| 19 |
+
|
| 20 |
+
| Average | ARC | HellaSwag | MMLU | TruthfulQA | Winogrande | GSM8K |
|
| 21 |
+
| --- | --- | --- | --- | --- | --- | --- |
|
| 22 |
+
| 67.33 | 59.64 | 81.82 | 61.69 | 53.23 | 78.45 | 69.14 |
|
| 23 |
+
|
| 24 |
+
For comparison the GSM8K score for the original `metamath/MetaMath-Mistral-7B` was 68.84 and average score was 65.78.
|