std::bad_alloc

[cindity]

Hi @lematt1991 ,
Thank you for your excellent work on this project.

I’ve received approval on Hugging Face and successfully set up the SAM Audio repository on my local machine with GPU support (23 GB VRAM). However, when running the following code, I encounter a std::bad_alloc error:

from sam_audio import SAMAudio, SAMAudioProcessor
import torchaudio

model = SAMAudio.from_pretrained("facebook/sam-audio-large")
processor = SAMAudioProcessor.from_pretrained("facebook/sam-audio-large")
model = model.eval().cuda()

# audio file path or torch tensor
file = "/code/fwd-local-scripts/pitchshift/new_voices/raw_mp3/my_voice.mp3"
description = ["A man speaking"]

batch = processor(
    audios=[file],
    descriptions=[description]).to("cuda")

result = model.separate(batch)

# Save separated audio
sample_rate = processor.audio_sampling_rate
torchaudio.save("target.wav", result.target.cpu(),
                sample_rate)      # The isolated sound
torchaudio.save("residual.wav", result.residual.cpu(),
                sample_rate)  # Everything else

root@9122383de6d5:/code/sam-audio# df -h /dev/shm
Filesystem      Size  Used Avail Use% Mounted on
shm              16G     0   16G   0% /dev/shm

root@9122383de6d5:/code/sam-audio# cat /sys/fs/cgroup/memory.max 2>/dev/null || cat /sys/fs/cgroup/memory/memory.limit_in_bytes
34359738368

root@9122383de6d5:/code/sam-audio# python tools/text_prompting.py 
terminate called after throwing an instance of 'std::bad_alloc'
  what():  std::bad_alloc
Aborted (core dumped)

Can you share how I can debug this?

[Eydahn8]

you’re basically the only one who got approved on Hugging Face, what’s the secret? We’re all still stuck on “pending”, and someone even got straight up rejected.

[xiejesses]

you’re basically the only one who got approved on Hugging Face, what’s the secret? We’re all still stuck on “pending”, and someone even got straight up rejected.

No, mine was also approved, and I am currently in the testing phase

[vladlearns]

you are preparing the batch, you are not out of vram

description = ["A man speaking"] - try to remove the brackets here

[cindity]

Hi @vladlearns,

Thank you for your instruction. However, I still have the same error after removing the brackets

[SkinnyLecttuce]

Try removing "result = model.separate(batch)" and add
"with torch.inference_mode(): result = model.separate(batch)"

[Avinash2236]

I am also facing the same error!!

[jetjodh]

Make sure your torchcodec version matches your torch version.

[Avinash2236]

Thanks changing the torchcodec version worked.
I think you meant "torchcodec" version and not "torchdecode".