Open AlpinDale opened 2 months ago
Hi @AlpinDale, sorry for the inconvenience. What type of hard-drive is it? (quite classic or a special mounted drive?). Asking because filelock doesn't always work properly on some filesystems. Independently from that, you can try to kill all huggingface_hub/hf_transfer processes and then run rm -rf /home/austin/.cache/huggingface/hub/.locks
to delete all current locks. This should fix your issues ( :crossed_fingers: ), though I can't explain why it happened in the first place.
Same issue here. Tried to delete the .locks
but it unfortunately didn't help. Instead, reducing the --max-workers
to something like 2 worked.
EG:
huggingface-cli download stabilityai/stable-diffusion-3.5-medium --max-workers 2
This is without using hf_transfer, and for a different model. In my case this did not hinder performance, but I imagine that varies much on your network speed.
EDIT: Spoke too soon. Didn't solve however reduced the frequency at least.
Describe the bug
I've been trying to download
NousResearch/Meta-Llama-3.1-8B-Instruct
with and withouthf-transfer
, but it consistently hangs at the 10GB point (2 shards with hf-transfer, half of each without), with this message being repeated every few seconds:Reproduction
Logs
System info