github/ColossalAI
1
0
Fork 0
mirror of https://github.com/hpcaitech/ColossalAI.git synced 2025-09-07 20:10:17 +00:00
Code Issues Packages Projects Releases Wiki Activity

[npu] change device to accelerator api (#5239)

Browse Source 
* update accelerator

* fix timer

* fix amp

* update

* fix

* update bug

* add error raise

* fix autocast

* fix set device

* remove doc accelerator

* update doc

* update doc

* update doc

* use nullcontext

* update cpu

* update null context

* change time limit for example

* udpate

* update

* update

* update

* [npu] polish accelerator code

---------

Co-authored-by: Xuanlei Zhao <xuanlei.zhao@gmail.com>
Co-authored-by: zxl <43881818+oahzxl@users.noreply.github.com>
This commit is contained in:
Hongxin Liu
2024-01-09 10:20:05 +08:00
committed by GitHub
parent dd2c28a323
commit d202cc28c0
128 changed files with 1773 additions and 868 deletions
Download Patch File Download Diff File Expand all files Collapse all files

7
colossalai/zero/gemini/gemini_ddp.py
View File

@@ -10,6 +10,7 @@ import torch.nn as nn
from torch.distributed import ProcessGroup
from torch.distributed.distributed_c10d import _get_default_group
from colossalai.accelerator import get_accelerator
from colossalai.checkpoint_io.utils import StateDictSharder, gather_distributed_param
from colossalai.interface import ModelWrapper
from colossalai.lazy import LazyTensor
@@ -27,7 +28,7 @@ from colossalai.tensor.d_tensor import (
is_distributed_tensor,
)
from colossalai.tensor.param_op_hook import ColoParamOpHookManager
from colossalai.utils import _cast_float, free_storage, get_current_device, is_ddp_ignored
from colossalai.utils import _cast_float, free_storage, is_ddp_ignored
from .chunk import Chunk, ChunkManager, TensorState, init_chunk_manager
from .gemini_hook import GeminiZeROHook
@@ -766,7 +767,7 @@ class GeminiDDP(ModelWrapper):
# move ignored parameters to CUDA
if is_ddp_ignored(p):
p.data = p.data.to(device=get_current_device(), dtype=self.mixed_precision)
p.data = p.data.to(device=get_accelerator().get_current_device(), dtype=self.mixed_precision)
continue
# create a fp16 parameter
@@ -815,7 +816,7 @@ class GeminiDDP(ModelWrapper):
for buffer in self.module.buffers():
if isinstance(buffer, LazyTensor):
buffer.materialize()
buffer.data = buffer.to(get_current_device())
buffer.data = buffer.to(get_accelerator().get_current_device())
if torch.is_floating_point(buffer):
buffer.data = buffer.to(self.mixed_precision)