Jacky Lee

Yeah.. tests are failing randomly and I’m not able to reproduce it locally. I suspect this is a problem with the CI runners. I’ll put this on hold for now.

> Oh yea...OOMs are segfaults in CI, maybe it's that? We probably should add an assert for this. > > If it's fast enough, we don't have to run the...

We might have a problem with OpenCL on Intel iGPUs. I'll try to reproduce the error on my Intel Mac. https://community.amd.com/t5/opencl/question-about-opencl-local-work-size-local-size-256-1-doesn-t/td-p/430396

I have to skip `test_speed_v_torch.py` due to the lack of VRAM in the Mac CI runner. Although we can probably add an option for `TINY=1` that reduces the size of...

Only `test_conv_with_bn` is failing now. It's likely memory related.

Closing this. Bye metal 🥲

Can you try running `GPU=1 python3 -m examples.whisper`? Edit: I guess you have an environment variable (METAL, CUDA, or GPU) set to x86_64. Can you check by running `printenv`?

Okay, try running `CPU=1 python3 -m examples.whisper`. Edit: We assume the device-related environment variables are either unset or set to an integer (usually 1).

Looks like we have a bug where the sentence keeps repeating. ``` 2277-149896-0004 How would the papers talk about it? How would the papers talk about it? How would the...

Added a comparison and similarity score. ``` -------------------------------------------------------------------------------------------------------------------------------- 2277-149896-0000 he was in a fevered state of mind owing to the blight his wifes action threatened to cast upon his entire...