palladium123

I benchmarked my RTX 3090 + 7920X using a 24 min video and beam_size=10: Medium: 117.4s, large 123.06s A 5% difference. For beam_size=5(default): Medium: 67.1s, large 91s (26.3% difference) For...

I created a PR for this in the past (before VAD was implemented), but I only implemented avg_logprob (sorry @guillaumekln typo there in my original PR) and no_speech_prob as I...

I tried quantizing the model with the workaround provided above, with no luck. I tried every permitted `format` which did not help. The error is (this one happened with `format...