[Misc] centralize all usage of environment variables (#4548)

This commit is contained in:
youkaichao
2024-05-02 11:13:25 -07:00
committed by GitHub
parent 1ff0c73a79
commit 5b8a7c1cb0
18 changed files with 220 additions and 64 deletions
@@ -1,10 +1,10 @@
import os
from contextlib import contextmanager
from typing import Any, List, Optional
import torch
import torch.distributed as dist
import vllm.envs as envs
from vllm.logger import init_logger
try:
@@ -54,9 +54,9 @@ def init_custom_ar() -> None:
return
# test nvlink first, this will filter out most of the cases
# where custom allreduce is not supported
if "CUDA_VISIBLE_DEVICES" in os.environ:
device_ids = list(
map(int, os.environ["CUDA_VISIBLE_DEVICES"].split(",")))
cuda_visible_devices = envs.CUDA_VISIBLE_DEVICES
if cuda_visible_devices:
device_ids = list(map(int, cuda_visible_devices.split(",")))
else:
device_ids = list(range(num_dev))
# this checks hardware and driver support for NVLink
+2 -2
View File
@@ -4,11 +4,11 @@
# Copyright (c) 2022, NVIDIA CORPORATION. All rights reserved.
"""Tensor and pipeline parallel groups."""
import contextlib
import os
from typing import Optional
import torch
import vllm.envs as envs
from vllm.logger import init_logger
logger = init_logger(__name__)
@@ -80,7 +80,7 @@ def init_distributed_environment(
# local_rank is not available in torch ProcessGroup,
# see https://github.com/pytorch/pytorch/issues/122816
if local_rank == -1 and distributed_init_method == "env://":
local_rank = int(os.environ['LOCAL_RANK'])
local_rank = envs.LOCAL_RANK
global _LOCAL_RANK
_LOCAL_RANK = local_rank
+5 -2
View File
@@ -9,6 +9,7 @@ from typing import Dict, Optional, Sequence
import torch
import torch.distributed as dist
import vllm.envs as envs
from vllm.logger import init_logger
from .parallel_state import get_cpu_world_group, get_local_rank
@@ -102,11 +103,13 @@ def gpu_p2p_access_check(i: int, j: int) -> bool:
is_distributed = dist.is_initialized()
num_dev = torch.cuda.device_count()
cuda_visible_devices = os.environ.get("CUDA_VISIBLE_DEVICES", None)
cuda_visible_devices = envs.CUDA_VISIBLE_DEVICES
if cuda_visible_devices is None:
cuda_visible_devices = ",".join(str(i) for i in range(num_dev))
VLLM_CONFIG_ROOT = envs.VLLM_CONFIG_ROOT
path = os.path.expanduser(
f"~/.config/vllm/gpu_p2p_access_cache_for_{cuda_visible_devices}.json")
f"{VLLM_CONFIG_ROOT}/vllm/gpu_p2p_access_cache_for_{cuda_visible_devices}.json"
)
os.makedirs(os.path.dirname(path), exist_ok=True)
if (not is_distributed or get_local_rank() == 0) \
and (not os.path.exists(path)):