Instruction Following Evaluation on Self-instruct Eval
56.7Win Rate (A)BPO
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| BPOBase LLM=claude-instant-1.2, Method A=BPO, Method B=Original, Evaluator=Claude-v1.32023.11 | 56.7 | 43.3 | |
| BPOBase LLM=gpt-3.5-turbo, Method A=BPO, Method B=Original, Evaluator=Claude-v1.32023.11 | 56.3 | 43.7 | |
| BPOBase LLM=text-bison, Method A=BPO, Method B=Original, Evaluator=Claude-v1.32023.11 | 56.3 | 43.7 | |
| BPOBase LLM=claude-2, Method A=BPO, Method B=Original, Evaluator=Claude-v1.32023.11 | 51.6 | 48.4 | |
| BPOBase LLM=gpt-4, Method A=BPO, Method B=Original, Evaluator=Claude-v1.32023.11 | 51.2 | 48.8 |