Vicuna 7B v1.1
LMSYS · 2023-04-12 · 6.7B parameters
Vicuna 7B v1.1 is LMSYS's autoregressive chat assistant, produced by supervised fine-tuning LLaMA 7B on about 70,000 user-shared conversations collected from ShareGPT. Version 1.1 refactored tokenization and the conversation separator to use the EOS token, improving stop-criterion handling and library compatibility, and fixed the supervised fine-tuning loss computation for better model quality.
Benchmark scores
| Benchmark | Score |
|---|---|
| EQ-Bench v2 | 26.1 |
| EQ-Bench v2 + MAGI-Hard Combined | 26.8 |
| MAGI-Hard | 27.4 |