Instruction Following on Alpaca instruction-following (test)
3.85PPLFull FT
Evaluation results
| Method | Links | |
|---|---|---|
| Full FTTraining Protocol=On Task A only, Stage=after Task A2026.02 | 3.85 | |
| Full FT SequentialTraining Protocol=Sequential (A then B), Stage=after Task A2026.02 | 3.85 | |
| LoRATraining Protocol=On Task A only, Stage=after Task A, Rank=82026.02 | 4.21 | |
| LoRA SequentialTraining Protocol=Sequential (A then B), Stage=after Task A, Rank=82026.02 | 4.21 | |
| LoRA A + LoRA BTraining Protocol=Separate adapters, Stage=after Task A, Rank=82026.02 | 4.21 | |
| LoRA A + LoRA BTraining Protocol=Separate adapters, Stage=after Task A+B, Rank=82026.02 | 4.21 | |
| NIWFStage=after Task A2026.02 | 4.35 | |
| NIWFStage=after Task A+B2026.02 | 4.35 | |
| LoRA SequentialTraining Protocol=Sequential (A then B), Stage=after Task A+B, Rank=82026.02 | 5.89 | |
| Full FT SequentialTraining Protocol=Sequential (A then B), Stage=after Task A+B2026.02 | 7.14 | |
| Base Mistral-7BTraining Protocol=None (No training), Stage=after Task A2026.02 | 8.92 | |
| Base Mistral-7BTraining Protocol=None (No training), Stage=after Task A+B2026.02 | 8.92 |