huggingface/text-generation-inference
5
0
Fork 0
mirror of https://github.com/huggingface/text-generation-inference.git synced 2025-09-09 03:14:53 +00:00
Code Issues Packages Projects Releases Wiki Activity

fix(server): Do not init process group if already initialized (#388)

Browse Source
This commit is contained in:
Antoni Baum 2023-06-26 03:32:54 -07:00 committed by GitHub
parent aefde28b45
commit ae466a8736
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
1 changed files with 13 additions and 8 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

21
server/text_generation_server/utils/dist.py
View File

@ -2,6 +2,7 @@ import os
import torch import torch
from datetime import timedelta from datetime import timedelta
from loguru import logger
class FakeBarrier: class FakeBarrier:
@ -59,13 +60,17 @@ def initialize_torch_distributed():
else: else:
if os.getenv("DEBUG", None) == "1": if os.getenv("DEBUG", None) == "1":
return FakeGroup(rank, world_size), rank, world_size return FakeGroup(rank, world_size), rank, world_size
# Call the init process.
torch.distributed.init_process_group( if not torch.distributed.is_initialized():
backend=backend, # Call the init process.
world_size=world_size, torch.distributed.init_process_group(
rank=rank, backend=backend,
timeout=timedelta(seconds=60), world_size=world_size,
pg_options=options, rank=rank,
) timeout=timedelta(seconds=60),
pg_options=options,
)
else:
logger.warning("torch.distributed is already initialized.")
return torch.distributed.group.WORLD, rank, world_size return torch.distributed.group.WORLD, rank, world_size