Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

LongformerForSequenceClassification训练速度慢 #1829

Open
LmYjQ opened this issue Nov 19, 2024 · 1 comment
Open

LongformerForSequenceClassification训练速度慢 #1829

LmYjQ opened this issue Nov 19, 2024 · 1 comment
Labels
ascend bug Something isn't working

Comments

@LmYjQ
Copy link

LmYjQ commented Nov 19, 2024

Describe the bug/ 问题描述 (Mandatory / 必填)
BertForSequenceClassification只要5分钟可以跑完一个epoch的数据集,LongformerForSequenceClassification预计要几十个小时。可以观察到NPU的HBM有占用,但是AIcore几乎没有占用。

两份代码区别只是定义tokenizer和from_pretrained的两行。

  • Hardware Environment(Ascend/GPU/CPU) / 硬件环境:
    ascend910b

  • Software Environment / 软件环境 (Mandatory / 必填):
    -- MindSpore version 2.3.1
    -- Python version 3.9.20
    -- mindnlp version 0.4.0
    -- OS platform and distribution (e.g., Linux Ubuntu 16.04): ky10.aarch64

因为代码在内网环境很难弄出来,正在尝试用启智平台复现。

@LmYjQ LmYjQ added the bug Something isn't working label Nov 19, 2024
@lvyufeng
Copy link
Collaborator

longformer涉及到循环,在Ascend上目前确实非常慢

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
ascend bug Something isn't working
Projects
None yet
Development

No branches or pull requests

2 participants