Narrow finetuning on Legal
85.9EM-F1TReFT
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| TReFTModel=Qwen-2.5-32B2026.06 | 85.9 | -0.3 | |
| Data interleavingModel=Qwen-2.5-32B2026.06 | 81.7 | -1.2 | |
| TReFTModel=GPT-OSS-20B2026.06 | 79.2 | -1.85 | |
| TReFTModel=Llama-3.1-8B2026.06 | 78.4 | -0.2 | |
| TReFTModel=Qwen-2.5-7B2026.06 | 77.7 | -0.7 | |
| Data interleavingModel=GPT-OSS-20B2026.06 | 75.1 | -2.25 | |
| Data interleavingModel=Qwen-2.5-7B2026.06 | 74.9 | -0.7 | |
| Data interleavingModel=Llama-3.1-8B2026.06 | 73 | -0.9 | |
| SFTModel=Llama-3.1-8B2026.06 | 61.4 | -0.9 | |
| SFTModel=GPT-OSS-20B2026.06 | 61.3 | -3.3 | |
| SFTModel=Qwen-2.5-32B2026.06 | 53.7 | -2.4 | |
| SFTModel=Qwen-2.5-7B2026.06 | 53.5 | -1.1 |