Speech Translation on ST-E2C
35.4BLEUFull KV
Evaluation results
| Method | Links | |
|---|---|---|
| Full KVModel=Qwen2.5-Omni-3B, KV-cache retention ratio=1.02026.04 | 35.4 | |
| AudioKVModel=Qwen2.5-Omni-3B, KV-cache retention ratio=0.82026.04 | 35.3 | |
| AudioKVModel=Qwen2.5-Omni-3B, KV-cache retention ratio=0.62026.04 | 35 | |
| AudioKVModel=Qwen2.5-Omni-3B, KV-cache retention ratio=0.42026.04 | 32.5 | |
| AudioKVModel=Gemma-3n-E2B, KV-cache retention ratio=0.82026.04 | 24.8 | |
| AudioKVModel=Gemma-3n-E2B, KV-cache retention ratio=0.62026.04 | 23.1 | |
| Full KVModel=Gemma-3n-E2B, KV-cache retention ratio=1.02026.04 | 22.8 | |
| AudioKVModel=Gemma-3n-E2B, KV-cache retention ratio=0.42026.04 | 18.9 |