Transformers: How convert pytorch to tf checkpoint?

Created on 19 Jan 2019 · 8Comments · Source: huggingface/transformers

How convert pytorch to tf checkpoint?

Help wanted wontfix

Source

semsevens

Most helpful comment

Thanks @nikitakit, do you think your scripts would make sense in the present repo as well or is it tided to your parsing application?

thomwolf on 28 Jan 2019

👍2

All 8 comments

I don't think such a conversion is currently implemented in this repository, but I have my own implementation here (if you're interested in adapting it for your use-case): https://github.com/nikitakit/self-attentive-parser/blob/8238e79e2089300db059eddff78229a09e254f70/export/export_bert.py#L94-L141

nikitakit on 21 Jan 2019

👍2

Thanks @nikitakit, do you think your scripts would make sense in the present repo as well or is it tided to your parsing application?

thomwolf on 28 Jan 2019

👍2

Closing this for now. Feel free to re-open.

thomwolf on 5 Feb 2019

Would be actually quite nice to have such a conversion script in order to serve pytorch models via bert-as-service

tholor on 7 Feb 2019

Any progress on this? As @tholor says, would be nice for bert-as-service. @nikitakit is it possible to run your script for any finetuned pytorch model? If so, any tips/suggestions on how to do that?

maxlund on 15 Mar 2019

Re-opening the issue.
I don't have time to work on that at the moment but I would be happy to welcome a PR if there is interest in this feature.

thomwolf on 15 Mar 2019

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.