github/ColossalAI
1
0
Fork 0
mirror of https://github.com/hpcaitech/ColossalAI.git synced 2025-09-10 13:30:19 +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

10
examples/language/bert/finetune.py
View File

@@ -18,11 +18,11 @@ from transformers import (
)
import colossalai
from colossalai.accelerator import get_accelerator
from colossalai.booster import Booster
from colossalai.booster.plugin import GeminiPlugin, HybridParallelPlugin, LowLevelZeroPlugin, TorchDDPPlugin
from colossalai.cluster import DistCoordinator
from colossalai.nn.optimizer import HybridAdam
from colossalai.utils import get_current_device
# ==============================
# Prepare Hyperparameters
@@ -59,7 +59,7 @@ def evaluate_model(
use_pipeline = isinstance(booster.plugin, HybridParallelPlugin) and booster.plugin.pp_size > 1
is_pp_last_stage = use_pipeline and booster.plugin.stage_manager.is_last_stage()
accum_loss = torch.zeros(1, device=get_current_device())
accum_loss = torch.zeros(1, device=get_accelerator().get_current_device())
for batch in dataloader:
batch = move_to_cuda(batch)
labels = batch["labels"]
@@ -88,8 +88,10 @@ def evaluate_model(
object_list = [None, None]
dist.broadcast_object_list(object_list, src=current_pp_group_ranks[-1], group=pp_group)
metric.add_batch(predictions=object_list[0].to(get_current_device()), references=labels)
accum_loss.add_(object_list[1].to(get_current_device()))
metric.add_batch(
predictions=object_list[0].to(get_accelerator().get_current_device()), references=labels
)
accum_loss.add_(object_list[1].to(get_accelerator().get_current_device()))
else:
batch = move_to_cuda(batch)