Instruction-following on TRACE
57.6AAPerTaskFT
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| PerTaskFTBackbone=LLaMA-2-7B-Chat2026.05 | 57.6 | — | |
| MTLBackbone=LLaMA-2-7B-Chat2026.05 | 52.3 | — | |
| Muon-OGDBackbone=LLaMA-2-7B-Chat2026.05 | 49.4 | -4.7 | |
| OSFTBackbone=LLaMA-2-7B-Chat2026.05 | 48.4 | -7.1 | |
| O-LoRABackbone=LLaMA-2-7B-Chat2026.05 | 41.3 | -6.2 | |
| SeqFTBackbone=LLaMA-2-7B-Chat2026.05 | 23 | -8.3 | |
| LoRASeqFTBackbone=LLaMA-2-7B-Chat2026.05 | 9.2 | -24.6 |