| { |
| "architecture": "whisper-causal", |
| "base_model": "openai/whisper-small.en", |
| "model": { |
| "n_mels": 80, |
| "d": 768, |
| "layers": 12, |
| "heads": 12, |
| "ffn": 3072, |
| "max_positions": 1500, |
| "n_out": 6, |
| "chunk": 1500, |
| "context": 750 |
| }, |
| "name": "cuelight-base", |
| "sample_rate": 16000, |
| "frame_samples": 320, |
| "outputs": [ |
| "p_speech", |
| "p_end_of_turn", |
| "aux_0", |
| "aux_1", |
| "aux_2", |
| "aux_3" |
| ], |
| "policies": { |
| "careful": { |
| "threshold": 0.58, |
| "min_silence_s": 0.7, |
| "timeout_s": 3.5 |
| }, |
| "snappy": { |
| "threshold": 0.66, |
| "min_silence_s": 0.4, |
| "timeout_s": 2.0 |
| } |
| }, |
| "policies_note": "careful: interrupts on ~1 in 20 pauses; snappy: ~1 in 10 (tuned on half of eot-bench English)" |
| } |
|
|
