Instruction Following on Human Instructions
0.22AccuracyBaseline
Evaluation results
| Method | Links | ||
|---|---|---|---|
| BaselineModel Backbone=Qwen2.5-3B, Training Config=normal SFT2026.05 | 0.22 | 9.3 | |
| BaselineModel Backbone=Llama-3.2-3B, Training Config=normal SFT2026.05 | 0.192 | 5.6 | |
| AsyncIOModel Backbone=Llama-3.2-3B, Training Config=Async-SFT2026.05 | 0.153 | 9.3 | |
| AsyncIOModel Backbone=Qwen2.5-3B, Training Config=Async-SFT2026.05 | 0.136 | 14.3 | |
| BaselineModel Backbone=Qwen2.5-3B, Training Config=no SFT2026.05 | 0.04 | — | |
| BaselineModel Backbone=Llama-3.2-3B, Training Config=no SFT2026.05 | 0.04 | — |