hf_text-generation-inference/server/text_generation_server/utils/dist.py

82 lines
2.3 KiB
Python
Raw Permalink Normal View History

2023-02-14 05:02:16 -07:00
import os
import torch
from datetime import timedelta
from loguru import logger
2023-02-14 05:02:16 -07:00
# Tensor Parallelism settings
RANK = int(os.getenv("RANK", "0"))
WORLD_SIZE = int(os.getenv("WORLD_SIZE", "1"))
# CUDA memory fraction
MEMORY_FRACTION = float(os.getenv("CUDA_MEMORY_FRACTION", "1.0"))
2023-02-14 05:02:16 -07:00
class FakeBarrier:
def wait(self):
pass
class FakeGroup:
def __init__(self, rank, size):
self._rank = rank
self._size = size
def allreduce(self, *args, **kwargs):
return FakeBarrier()
def allgather(self, inputs, local_tensor, **kwargs):
assert (
len(inputs[0]) == len(local_tensor) == 1
), f"{len(inputs[0])} != {len(local_tensor)} != 1, and the FakeGroup is supposed to join on simple tensors"
for input_ in inputs:
input_[0].data = local_tensor[0].data
return FakeBarrier()
def barrier(self, *args, **kwargs):
return FakeBarrier()
def size(self):
return self._size
def rank(self):
return self._rank
2023-02-14 05:02:16 -07:00
def initialize_torch_distributed():
if torch.cuda.is_available():
from torch.distributed import ProcessGroupNCCL
# Set the device id.
assert WORLD_SIZE <= torch.cuda.device_count(), "Each process is one gpu"
device = RANK % torch.cuda.device_count()
2023-02-14 05:02:16 -07:00
torch.cuda.set_device(device)
torch.cuda.set_per_process_memory_fraction(MEMORY_FRACTION, device)
2023-02-14 05:02:16 -07:00
backend = "nccl"
options = ProcessGroupNCCL.Options()
options.is_high_priority_stream = True
options._timeout = timedelta(seconds=60)
else:
backend = "gloo"
options = None
if WORLD_SIZE == 1:
return FakeGroup(RANK, WORLD_SIZE), RANK, WORLD_SIZE
else:
if os.getenv("DEBUG", None) == "1":
return FakeGroup(RANK, WORLD_SIZE), RANK, WORLD_SIZE
if not torch.distributed.is_initialized():
# Call the init process.
torch.distributed.init_process_group(
backend=backend,
world_size=WORLD_SIZE,
rank=RANK,
timeout=timedelta(seconds=60),
pg_options=options,
)
else:
logger.warning("torch.distributed is already initialized.")
2023-02-14 05:02:16 -07:00
return torch.distributed.group.WORLD, RANK, WORLD_SIZE