Frankenmodels
Collection
They're not supposed to be that size! Neat, right?
β’
8 items
β’
Updated
β’
3
Similar to llama2-22b, but with BLOCK_DIAGONAL=false in the merge and twice the fine-tuning tokens.
Again, not intended for direct use - meant as a base for further tuning and merging.
Detailed results can be found here
Metric | Value |
---|---|
Avg. | 46.86 |
ARC (25-shot) | 58.28 |
HellaSwag (10-shot) | 82.69 |
MMLU (5-shot) | 54.53 |
TruthfulQA (0-shot) | 39.23 |
Winogrande (5-shot) | 75.93 |
GSM8K (5-shot) | 11.22 |
DROP (3-shot) | 6.17 |