The LLaMA based Pygmalion-7b model:
https://huggingface.co/PygmalionAI/pygmalion-7b
Merged alongside Tloen's Alpaca LoRA:
https://huggingface.co/tloen/alpaca-lora-7b
This was done to test whether LoRAs trained for other LLaMA fine tunes worked for Pygmalion, and to have it available on inference backends that do not support LoRAs just yet.
Treat this as a normal HF Transformers model.
Open LLM Leaderboard Evaluation Results
Detailed results can be found here
| Metric | Value |
|---|---|
| Avg. | 44.27 |
| ARC (25-shot) | 53.24 |
| HellaSwag (10-shot) | 76.92 |
| MMLU (5-shot) | 35.92 |
| TruthfulQA (0-shot) | 39.44 |
| Winogrande (5-shot) | 72.22 |
| GSM8K (5-shot) | 1.21 |
| DROP (3-shot) | 30.91 |