Spaces:
Running
on
Zero
Running
on
Zero
File size: 2,320 Bytes
42f2c22 e276ced 42f2c22 e276ced 42f2c22 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 |
# // Copyright (c) 2025 Bytedance Ltd. and/or its affiliates
# //
# // Licensed under the Apache License, Version 2.0 (the "License");
# // you may not use this file except in compliance with the License.
# // You may obtain a copy of the License at
# //
# // http://www.apache.org/licenses/LICENSE-2.0
# //
# // Unless required by applicable law or agreed to in writing, software
# // distributed under the License is distributed on an "AS IS" BASIS,
# // WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# // See the License for the specific language governing permissions and
# // limitations under the License.
"""
Distributed basic functions.
"""
import os
from datetime import timedelta
import torch
import torch.distributed as dist
from torch.nn.parallel import DistributedDataParallel
def get_global_rank() -> int:
"""
Get the global rank, the global index of the GPU.
"""
return int(os.environ.get("RANK", "0"))
def get_local_rank() -> int:
"""
Get the local rank, the local index of the GPU.
"""
return int(os.environ.get("LOCAL_RANK", "0"))
def get_world_size() -> int:
"""
Get the world size, the total amount of GPUs.
"""
return int(os.environ.get("WORLD_SIZE", "1"))
def get_device() -> torch.device:
"""
Get current rank device.
"""
return torch.device("cuda", get_local_rank())
def barrier_if_distributed(*args, **kwargs):
"""
Synchronizes all processes if under distributed context.
"""
if dist.is_initialized():
return dist.barrier(*args, **kwargs)
def init_torch(cudnn_benchmark=True, timeout=timedelta(seconds=600)):
"""
Common PyTorch initialization configuration.
"""
torch.backends.cuda.matmul.allow_tf32 = True
torch.backends.cudnn.allow_tf32 = True
torch.backends.cudnn.benchmark = cudnn_benchmark
torch.cuda.set_device(get_local_rank())
dist.init_process_group(
backend="nccl",
rank=get_global_rank(),
world_size=get_world_size(),
timeout=timeout,
)
def convert_to_ddp(module: torch.nn.Module, **kwargs) -> DistributedDataParallel:
return DistributedDataParallel(
module=module,
device_ids=[get_local_rank()],
output_device=get_local_rank(),
**kwargs,
)
|