Explanatory Reasoning on Explanatory Reason
65.66BLEUQwen2.5-0.5B (Base model)
Evaluation Results
| Method | Links | |
|---|---|---|
| Qwen2.5-0.5B (Base model)Injection=Base model, Train Setup=Zero-shot2026.01 | 65.66 | |
| VAE (query_add)Injection=query_add, Train Setup=prior=False2026.01 | 64.49 | |
| VAE (kv_mem)Injection=kv_mem, Train Setup=prior=False2026.01 | 63.13 | |
| VAE (query_add)Injection=query_add, Train Setup=weight=0.52026.01 | 62.2 | |
| VAE (kv_add)Injection=kv_add, Train Setup=prior=False2026.01 | 61.27 | |
| VAE (query_add)Injection=query_add, Train Setup=weight=1.02026.01 | 61.18 | |
| VAE (kv_mem)Injection=kv_mem, Train Setup=weight=1.02026.01 | 60.22 | |
| VAE (kv_mem)Injection=kv_mem, Train Setup=weight=0.52026.01 | 59.87 | |
| VAE (kv_add)Injection=kv_add, Train Setup=weight=0.52026.01 | 26.23 | |
| VAE (kv_add)Injection=kv_add, Train Setup=weight=1.02026.01 | 20.49 | |
| Qwen2.5-0.5B (Base model)Injection=Base model, Train Setup=Few-shot (FT)2026.01 | 8.01 | |
| Qwen2.5-0.5B (Base model)Injection=Base model, Train Setup=Few-shot2026.01 | 7.65 |