slower inference with 1.0.0 compared to 0.10.0

[thomasmol]

I am getting about 4x slower inference with version 1.0.0 compared to 0.10.0.

I used a 1 min 17 sec sample file, run on an A10G GPU

Before: Finished transcribing, took 5.043 seconds. Detected language: en

Now: Finished transcribing, took 21.936 seconds. Detected language: en

I did not change my config other than upgrading to 1.0.0

Only seems to happen on very short files, tested a 25 min file that transcribed at the same speed as before.

Any ideas?

[trungkienbkhn]

@thomasmol , hello. Can you check this PR and rerun your tests ?

[thomasmol]

That fixes it! Finished transcribing, took 5.1937 seconds. Detected language: en back to normal