wsl (windows subsidiary linux) debian install ERROR: Could not install packages due to an OSError: [Errno 28] No space left on device

[sugizo]

steps on windows

wsl --install debian
wsl

inside linux

sudo apt update
sudo apt install -y python3-venv

sudo apt autoremove
sudo apt clean

cd /mnt/c/Users/yoshiki/Downloads/

python3 -m venv ./openai-whisper
source ./openai-whisper/bin/activate

pip cache purge

pip install --no-cache-dir openai-whisper

things already tried

(openai-whisper) yoshiki@LAPTOP-AEUH1H9T:/mnt/c$ df -h /
Filesystem      Size  Used Avail Use% Mounted on
/dev/sdd       1007G  1.1G  955G   1% /

(openai-whisper) yoshiki@LAPTOP-AEUH1H9T:/mnt/c$ sudo fstrim -av
[sudo] password for yoshiki:
/: 103.9 MiB (108924928 bytes) trimmed on /dev/sdd

(openai-whisper) yoshiki@LAPTOP-AEUH1H9T:/mnt/c$ df -h /
Filesystem      Size  Used Avail Use% Mounted on
/dev/sdd       1007G  1.1G  955G   1% /

check diskmgmt.msc on windows
still have freespace around 37 gb free (16 % free)

question
any way out to solve this problem ?

thanks and best regards

[harshagarwalnyu]

Couple of things, and that [Errno 28] is misleading here. Your / has 955G free (you showed it), and WSL's /tmp isn't a RAM disk, it lives on the ext4 VHDX, so this isn't a memory thing and bumping memory in .wslconfig won't change anything.

The detail that matters is where the venv is:

cd /mnt/c/Users/yoshiki/Downloads/
python3 -m venv ./openai-whisper

/mnt/c is your Windows C: drive mounted into WSL, and you said that drive has about 37 GB free. openai-whisper installs torch, and the default torch build drags the whole NVIDIA CUDA stack with it (nvidia-cublas, nvidia-cudnn, nvidia-cuda-runtime, triton, and friends), which unpacks to a lot. pip writes all of that onto the Windows drive because that's where the venv is, so that's the disk that fills up, not /.

Two fixes, either works:

1. Put the venv on the Linux side where you have 955G:

cd ~
python3 -m venv openai-whisper
source openai-whisper/bin/activate
pip install --no-cache-dir openai-whisper

2. If you don't need GPU, install CPU-only torch first so you skip the multi-GB CUDA packages:

pip install torch --index-url https://download.pytorch.org/whl/cpu
pip install --no-cache-dir openai-whisper

Building under ~ instead of /mnt/c is also a lot faster, the Windows mount is slow for thousands of small files.

[sugizo]

thanks for pointer

another way around :

mkdir -p ~/pip_cache
export TMPDIR=~/pip_cache
pip install openai-whisper

thanks and best regards