[CI] Disable non-lazy string operation on logging (#4326)

Co-authored-by: Danny Guinther <dguinther@neuralmagic.com>
This commit is contained in:
SangBin Cho
2024-04-26 00:16:58 -07:00
committed by GitHub
co-authored by Danny Guinther
parent 2f30e7c72f
commit a88081bf76
31 changed files with 176 additions and 149 deletions
@@ -37,7 +37,7 @@ def init_custom_ar() -> None:
return
if world_size not in _SUPPORTED_WORLD_SIZES:
logger.warn(
logger.warning(
"Custom allreduce is disabled due to an unsupported world size: "
"%d. Supported world sizes: %s. To silence this warning, specify"
" disable_custom_all_reduce=True explicitly.", world_size,
@@ -47,7 +47,7 @@ def init_custom_ar() -> None:
# note: num dev can be larger than world_size if we're only using
# first few GPUs
if num_dev < world_size:
logger.warn(
logger.warning(
"Cannot test GPU P2P because not all GPUs are visible to the "
"current process. This might be the case if 'CUDA_VISIBLE_DEVICES'"
" is set.")
@@ -62,7 +62,7 @@ def init_custom_ar() -> None:
# this checks hardware and driver support for NVLink
full_nvlink = _is_full_nvlink(device_ids)
if world_size > 2 and not full_nvlink:
logger.warn(
logger.warning(
"Custom allreduce is disabled because it's not supported on more"
" than two PCIe-only GPUs. To silence this warning, specify"
" disable_custom_all_reduce=True explicitly.")
@@ -71,7 +71,7 @@ def init_custom_ar() -> None:
# this is expensive to compute at the first time
# then we cache the result
if not _can_p2p(rank, world_size):
logger.warn(
logger.warning(
"Custom allreduce is disabled because your platform lacks GPU P2P"
" capability or P2P test failed. To silence this warning, specify"
" disable_custom_all_reduce=True explicitly.")
@@ -43,15 +43,16 @@ try:
nccl = ctypes.CDLL(so_file)
except Exception as e:
logger.error(
f"Failed to load NCCL library from {so_file} ."
"Failed to load NCCL library from %s ."
"It is expected if you are not running on NVIDIA/AMD GPUs."
"Otherwise, the nccl library might not exist, be corrupted "
f"or it does not support the current platform {platform.platform()}."
f"One solution is to download libnccl2 version 2.18 from "
f"https://developer.download.nvidia.com/compute/cuda/repos/ "
f"and extract the libnccl.so.2 file. If you already have the "
f"library, please set the environment variable VLLM_NCCL_SO_PATH"
" to point to the correct nccl library path.")
"or it does not support the current platform %s."
"One solution is to download libnccl2 version 2.18 from "
"https://developer.download.nvidia.com/compute/cuda/repos/ "
"and extract the libnccl.so.2 file. If you already have the "
"library, please set the environment variable VLLM_NCCL_SO_PATH"
" to point to the correct nccl library path.", so_file,
platform.platform())
raise e
# === export types and functions from nccl to Python ===
@@ -14,7 +14,7 @@ try:
except Exception as e:
# in non-NVIDIA environments, we can't import the nccl module
# e.g. when running on machines with AMD GPUs
logger.info(f"Failed to import NCCL library: {e}")
logger.info("Failed to import NCCL library: %s", e)
logger.info("It is expected if you are not running on NVIDIA GPUs.")
pass
@@ -40,7 +40,7 @@ def set_pynccl_stream(stream: torch.cuda.Stream):
def init_process_group(group: Optional[ProcessGroup] = None) -> None:
assert not is_initialized()
global comm
logger.info(f"vLLM is using nccl=={ncclGetVersion()}")
logger.info("vLLM is using nccl==%s", ncclGetVersion())
comm = NCCLCommunicator(group=group)
+4 -2
View File
@@ -57,8 +57,10 @@ def init_distributed_environment(
local_rank: int = -1,
backend: str = "nccl",
):
logger.debug(f"{world_size=} {rank=} {local_rank=} "
f"{distributed_init_method=} {backend=}")
logger.debug(
"world_size=%d rank=%d local_rank=%d "
"distributed_init_method=%s backend=%s", world_size, rank, local_rank,
distributed_init_method, backend)
if not torch.distributed.is_initialized():
assert distributed_init_method is not None, (
"distributed_init_method must be provided when initializing "
+2 -2
View File
@@ -112,7 +112,7 @@ def gpu_p2p_access_check(i: int, j: int) -> bool:
and (not os.path.exists(path)):
# only the local master process (with local_rank == 0) can
# enter this block to calculate the cache
logger.info(f"generating GPU P2P access cache for in {path}")
logger.info("generating GPU P2P access cache for in %s", path)
cache = {}
for _i in range(num_dev):
for _j in range(num_dev):
@@ -126,7 +126,7 @@ def gpu_p2p_access_check(i: int, j: int) -> bool:
if is_distributed:
cpu_world_group = get_cpu_world_group()
dist.barrier(cpu_world_group)
logger.info(f"reading GPU P2P access cache from {path}")
logger.info("reading GPU P2P access cache from %s", path)
with open(path, "r") as f:
cache = json.load(f)
_gpu_p2p_access_cache = cache