github/ColossalAI
1
0
Fork 0
mirror of https://github.com/hpcaitech/ColossalAI.git synced 2025-09-16 14:41:53 +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

6
colossalai/kernel/extensions/flash_attention/utils.py
View File

@@ -6,7 +6,7 @@ import torch
import torch.nn.functional as F
from einops import rearrange
from colossalai.utils.device import get_current_device
from colossalai.accelerator import get_accelerator
class Unpad(torch.autograd.Function):
@@ -70,7 +70,9 @@ class SeqLenInfo:
cu_seqlens: torch.Tensor = None
@staticmethod
def materialize(attn_mask: torch.Tensor = None, size: Tuple[int] = None, device=get_current_device()):
def materialize(
attn_mask: torch.Tensor = None, size: Tuple[int] = None, device=get_accelerator().get_current_device()
):
if attn_mask is not None:
indices = torch.nonzero(attn_mask.flatten(), as_tuple=False).flatten().to(device)
seqlens = attn_mask.sum(dim=-1, dtype=torch.int32).flatten()