license: apache-2.0
datasets:
- abacusai/MetaMathFewshot
- shahules786/orca-chat
- anon8231489123/ShareGPT_Vicuna_unfiltered
Trained on the MetamathFewshot (https://huggingface.co/datasets/abacusai/MetaMathFewshot) dataset from base Mistral, as well as the Vicuna (https://huggingface.co/datasets/anon8231489123/ShareGPT_Vicuna_unfiltered) dataset and the OrcaChat (https://huggingface.co/datasets/shahules786/orca-chat) dataset.
Instruction tuned with the following parameters:
- LORA, Rank 8, Alpha 16, Dropout 0.05, all modules (QKV and MLP)
- 3 epochs
- Micro Batch Size 32 over 4xH100, gradient accumulation steps = 1
- AdamW with learning rate 5e-5
Evaluation Results
HuggingFace Leaderboard
| Average | ARC | HellaSwag | MMLU | TruthfulQA | Winogrande | GSM8K |
|---|---|---|---|---|---|---|
| 67.33 | 59.64 | 81.82 | 61.69 | 53.23 | 78.45 | 69.14 |
For comparison the GSM8K score for the original metamath/MetaMath-Mistral-7B was 68.84 and average score was 65.78.
MT-Bench
########## First turn ##########
score
model turn
fewshot_metamath_orcavicuna_mistral 1 6.9
########## Second turn ##########
score
model turn
fewshot_metamath_orcavicuna_mistral 2 6.51875
########## Average ##########
score
model
fewshot_metamath_orcavicuna_mistral 6.709375