github/ColossalAI
1
0
Fork 0
mirror of https://github.com/hpcaitech/ColossalAI.git synced 2025-09-08 20:40:34 +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

8
colossalai/legacy/nn/layer/parallel_2d/_operation.py
View File

@@ -5,10 +5,10 @@ import torch.distributed as dist
from torch import Tensor
from torch.cuda.amp import custom_bwd, custom_fwd
from colossalai.accelerator import get_accelerator
from colossalai.legacy.communication.collective import all_gather, all_reduce, reduce_scatter
from colossalai.legacy.context.parallel_mode import ParallelMode
from colossalai.legacy.core import global_context as gpc
from colossalai.utils import get_current_device
def matmul_2d(
@@ -250,7 +250,7 @@ class Matmul_AB_2D(torch.autograd.Function):
B_shape = B.shape
B = B.reshape((-1, B_shape[-1]))
C_shape = (A.shape[0], B.shape[-1])
C = torch.zeros(C_shape, dtype=A.dtype, device=get_current_device())
C = torch.zeros(C_shape, dtype=A.dtype, device=get_accelerator().get_current_device())
# use circular buffer to store the communication tensor
# 2 is enough for all cases
@@ -399,7 +399,7 @@ class Matmul_ABT_2D(torch.autograd.Function):
B_shape = B.shape
B = B.reshape((-1, B_shape[-1]))
C_shape = (A.shape[0], B.shape[0])
C = torch.empty(C_shape, dtype=A.dtype, device=get_current_device())
C = torch.empty(C_shape, dtype=A.dtype, device=get_accelerator().get_current_device())
# use circular buffer to store the communication tensor
# 2 is enough for all cases
@@ -556,7 +556,7 @@ class Matmul_ATB_2D(torch.autograd.Function):
B_shape = B.shape
B = B.reshape((-1, B_shape[-1]))
C_shape = (A.shape[-1], B.shape[-1])
C = torch.empty(C_shape, dtype=A.dtype, device=get_current_device())
C = torch.empty(C_shape, dtype=A.dtype, device=get_accelerator().get_current_device())
# use circular buffer to store the communication tensor
# 2 is enough for all cases