You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
I replaced the server Makefile to serve gpt2 without sharding like SAFETENSORS_FAST_GPU=1 python -m torch.distributed.run --nproc_per_node=2 text_generation_server/cli.py serve gpt2
Yesterday, the server and router worked fine, but the next time when I restarted the server, it doesn't work.
The issue I'm facing
The text was updated successfully, but these errors were encountered:
You don't need to launch using torch.distributed.run when you are not sharding your model.
You can directly use text-generation-server serve gpt2.
Since you are running with --nproc_per_node=2, the two process are competing for the same address.
I replaced the server Makefile to serve gpt2 without sharding like
SAFETENSORS_FAST_GPU=1 python -m torch.distributed.run --nproc_per_node=2 text_generation_server/cli.py serve gpt2
Yesterday, the server and router worked fine, but the next time when I restarted the server, it doesn't work.
The issue I'm facing
The text was updated successfully, but these errors were encountered: