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

tensorflow版本与pytorch版本相差多大呢? #18

Closed
luhua-rain opened this issue Aug 27, 2019 · 5 comments
Closed

tensorflow版本与pytorch版本相差多大呢? #18

luhua-rain opened this issue Aug 27, 2019 · 5 comments

Comments

@luhua-rain
Copy link

你好!
非常感谢开源模型!
但在线下我自己测试的时候,发现pytorch的效果要比tensorflow的小很多。请问:你们有测试过pytorch版本的效果吗?或者是我自己的代码有问题?

@ymcui
Copy link
Owner

ymcui commented Aug 28, 2019

我没有测过PyTorch版本,仅使用了Huggingface最新版的转换脚本。
效果以TF版本为准,你可以自行将TF权重转换为PT版本,确保转换无误。

@FrankWork
Copy link

@basketballandlearn 请问你测试的差距有多大?

@lsq357
Copy link

lsq357 commented Sep 1, 2019

@basketballandlearn 请问这个问题解决了吗?

  1. 我用keras-xlnet(https://github.com/CyberZHG/keras-xlnet) 跑数据集LCQMC,一直收敛不了。。
  2. 我用keras-xlnet跑实体识别,我自己训练的xlnet-base,只比bert-base差几个千分点而已,但是用这个,差了几个百分点(8个到9个)

@brightmart
Copy link

@basketballandlearn @FrankWork 你们后来有什么测试、发现或解决方法了吗?
LCQMC是中文的。尝试着训练了一个XLNet_Large中文版,超大数据量,训练完了(累计见了1亿元个训练数据),结果在LCQMC验证集的准确率比其他模型差了近10个点。

Best result | eval_accuracy 0.802749693394 | eval_loss 0.492308169603 | global_step 1000 | loss 0.481289118528

你们有尝试过在其他任务上的效果吗?

@luhua-rain
Copy link
Author

其他任务都效果都比bert要好几个点,我试过许多任务,没有低过bert的

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

5 participants