Instruction Following Evaluation on Dolly Eval
62A Win RateBPO
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| BPOBase LLM=gpt-4, Method A=BPO, Method B=Original, Evaluator=Claude-v1.32023.11 | 62 | 38 | |
| BPOBase LLM=text-bison, Method A=BPO, Method B=Original, Evaluator=Claude-v1.32023.11 | 60.5 | 39.5 | |
| BPOBase LLM=gpt-3.5-turbo, Method A=BPO, Method B=Original, Evaluator=Claude-v1.32023.11 | 60 | 40 | |
| BPOBase LLM=claude-instant-1.2, Method A=BPO, Method B=Original, Evaluator=Claude-v1.32023.11 | 51.5 | 48.5 | |
| BPOBase LLM=claude-2, Method A=BPO, Method B=Original, Evaluator=Claude-v1.32023.11 | 50.5 | 49.5 |