ArkaAbacus
commited on
Commit
•
f557d97
1
Parent(s):
10f189d
Update README.md
Browse files
README.md
CHANGED
@@ -17,10 +17,29 @@ Instruction tuned with the following parameters:
|
|
17 |
- Micro Batch Size 32 over 4xH100, gradient accumulation steps = 1
|
18 |
- AdamW with learning rate 5e-5
|
19 |
|
20 |
-
|
|
|
|
|
21 |
|
22 |
| Average | ARC | HellaSwag | MMLU | TruthfulQA | Winogrande | GSM8K |
|
23 |
| --- | --- | --- | --- | --- | --- | --- |
|
24 |
| 67.33 | 59.64 | 81.82 | 61.69 | 53.23 | 78.45 | 69.14 |
|
25 |
|
26 |
-
For comparison the GSM8K score for the original `metamath/MetaMath-Mistral-7B` was 68.84 and average score was 65.78.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
17 |
- Micro Batch Size 32 over 4xH100, gradient accumulation steps = 1
|
18 |
- AdamW with learning rate 5e-5
|
19 |
|
20 |
+
## Evaluation Results
|
21 |
+
|
22 |
+
### HuggingFace Leaderboard
|
23 |
|
24 |
| Average | ARC | HellaSwag | MMLU | TruthfulQA | Winogrande | GSM8K |
|
25 |
| --- | --- | --- | --- | --- | --- | --- |
|
26 |
| 67.33 | 59.64 | 81.82 | 61.69 | 53.23 | 78.45 | 69.14 |
|
27 |
|
28 |
+
For comparison the GSM8K score for the original `metamath/MetaMath-Mistral-7B` was 68.84 and average score was 65.78.
|
29 |
+
|
30 |
+
### MT-Bench
|
31 |
+
|
32 |
+
########## First turn ##########
|
33 |
+
score
|
34 |
+
model turn
|
35 |
+
fewshot_metamath_orcavicuna_mistral 1 6.9
|
36 |
+
|
37 |
+
########## Second turn ##########
|
38 |
+
score
|
39 |
+
model turn
|
40 |
+
fewshot_metamath_orcavicuna_mistral 2 6.51875
|
41 |
+
|
42 |
+
########## Average ##########
|
43 |
+
score
|
44 |
+
model
|
45 |
+
fewshot_metamath_orcavicuna_mistral 6.709375
|