Automatic Speech Recognition on ASR EN
95WARAudioKV
Evaluation results
| Method | Links | |
|---|---|---|
| AudioKVModel=Qwen2.5-Omni-3B, KV-cache retention ratio=0.82026.04 | 95 | |
| Full KVModel=Qwen2.5-Omni-3B, KV-cache retention ratio=1.02026.04 | 94.7 | |
| AudioKVModel=Qwen2.5-Omni-3B, KV-cache retention ratio=0.62026.04 | 93.9 | |
| Full KVModel=Gemma-3n-E4B, KV-cache retention ratio=1.02026.04 | 92.6 | |
| AudioKVModel=Gemma-3n-E4B, KV-cache retention ratio=0.82026.04 | 92.6 | |
| AudioKVModel=Gemma-3n-E4B, KV-cache retention ratio=0.62026.04 | 91.7 | |
| AudioKVModel=Qwen2.5-Omni-3B, KV-cache retention ratio=0.42026.04 | 90 | |
| AudioKVModel=Gemma-3n-E4B, KV-cache retention ratio=0.42026.04 | 80.8 |