Generative Task on PubMedQA labeled (0.8:0.2)
76.8AccuracyLoRA + SAM
Evaluation Results
| Method | Links | |
|---|---|---|
| LoRA + SAMBackbone=Ministral-14B2026.06 | 76.8 | |
| LoRA + NoisyTuneBackbone=Ministral-14B2026.06 | 76.7 | |
| GRAINBackbone=Qwen-14B2026.06 | 76.58 | |
| LoRA + PCGradBackbone=Qwen-14B2026.06 | 76.1 | |
| LoRA + SWABackbone=Ministral-14B2026.06 | 75.95 | |
| LoRA + SAMBackbone=Qwen-14B2026.06 | 75.65 | |
| GRAINBackbone=Ministral-14B2026.06 | 75.2 | |
| LoRA + FocalLossBackbone=Ministral-14B2026.06 | 74.66 | |
| LoRABackbone=Qwen-14B2026.06 | 74.58 | |
| LoRA + NoisyTuneBackbone=Qwen-14B2026.06 | 74.2 | |
| GRAINBackbone=Qwen-7B2026.06 | 74.15 | |
| LoRA + FocalLossBackbone=Qwen-14B2026.06 | 73.8 | |
| LoRA + SWABackbone=Qwen-14B2026.06 | 73.65 | |
| LoRA + PCGradBackbone=Ministral-14B2026.06 | 73.15 | |
| LoRABackbone=Qwen-7B2026.06 | 72.6 | |
| LoRA + FocalLossBackbone=Qwen-7B2026.06 | 71.62 | |
| LoRA + SAMBackbone=Qwen-7B2026.06 | 70.95 | |
| LoRA + NoisyTuneBackbone=Qwen-7B2026.06 | 70.05 | |
| LoRA + PCGradBackbone=Mistral-7B2026.06 | 69.9 | |
| LoRA + SWABackbone=Qwen-7B2026.06 | 68.15 | |
| LoRABackbone=Ministral-14B2026.06 | 67.9 | |
| GRAINBackbone=Mistral-7B2026.06 | 67.6 | |
| LoRA + SWABackbone=Mistral-7B2026.06 | 67.55 | |
| LoRA + SAMBackbone=Mistral-7B2026.06 | 66.95 | |
| LoRA + PCGradBackbone=Qwen-7B2026.06 | 66.4 | |
| LoRA + NoisyTuneBackbone=Mistral-7B2026.06 | 64.86 | |
| LoRABackbone=Mistral-7B2026.06 | 46.05 | |
| LoRA + FocalLossBackbone=Mistral-7B2026.06 | 20.6 |