-
Notifications
You must be signed in to change notification settings - Fork 353
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Whisper JAX is not faster than Whisper in colab GPU environment. #152
Comments
I have same question |
not only on colab but also on consumer hardware. I am able to run the whisper medium on my 8 VRAM GPU with no issue but using Whisper Jax i have no idea why i need to run it in dtype float16 to do not end up with OOM error. Is there a logic explanation for this ? |
Hey @bianxg - it looks like you're measuring the compilation time, which is supposed to be slow. Any subsequent calls to the pipeline will be much faster since we leverage the compiled function. You can see this in action in this Kaggle notebook: https://www.kaggle.com/code/sgandhi99/whisper-jax-tpu |
@bianxg @r2d209git @WasamiKirua Has any one had luck getting better results? @sanchit-gandhi Any tips on how that was optimized on the demo? I'm at around 11s where It feels like 1-3 seconds on the huggingface demo you have. Some benchmarks on my side:
As you can see same audio, same post processing time, huge transcription boost! Help on how to lower that 12seconds is much appreciated. |
Whisper JAX is not faster than Whisper in colab T4 GPU environment. Why?
I tested with a 841 seconds long audio file. The Whisper JAX used 182 seconds and Whisper used only 148 seconds.( Both use small model)
Please reference the Whisper JAX test code:
https://drive.google.com/file/d/1T9sGsOS4md5169jAnSpQX_tHGbS4yFEC/view?usp=sharing
The text was updated successfully, but these errors were encountered: