A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 17, 03:00 UTC
[Core][Distributed] improve logging for init dist (#4042)
youkaichao committed
Apr 13, 2024 at 07:12 UTC
98afde19fc273b1e6a695990b93ec07157b856f1
1 file changed
+6
vllm/distributed/parallel_state.py
+6
@@ -8,6 +8,10 @@ from typing import Optional
8
9
import torch
10
11
+from vllm.logger import init_logger
12
+
13
+logger = init_logger(__name__)
14
+
15
# Tensor model parallel group that the current rank belongs to.
16
_TENSOR_MODEL_PARALLEL_GROUP = None
17
# Pipeline model parallel group that the current rank belongs to.
@@ -45,6 +49,8 @@ def init_distributed_environment(
49
local_rank: int = -1,
50
backend: str = "nccl",
51
):
52
+ logger.debug(f"{world_size=} {rank=} {local_rank=} "
53
+ f"{distributed_init_method=} {backend=}")
54
if not torch.distributed.is_initialized():
55
assert distributed_init_method is not None, (
56
"distributed_init_method must be provided when initializing "