Forwarded from Feed-Master
doing something
Discovered it: https://github.com/pytorch/pytorch/issues/134084 #ai #torch
Can be configured using:
or a env var TORCHINDUCTOR_COMPILE_THREADS
torch._inductor.config.worker_start_method = 'spawn'
torch._inductor.config.compile_threads = 2
or a env var TORCHINDUCTOR_COMPILE_THREADS
doing something
https://github.com/rustprooflabs/pgdd #pgsql #rust
tg_image_3284352399.png
60.8 KB
so easy to integrate and use
doing something
Whisper sometimes generates FloatTensor and batch_decode can't do own job #ai #asr #whisper
dtype must be
torch.int64 , but it generates torch.float32