'Embedding' object has no attribute 'device' for MPT #1082
Comments
|
Detail: |
junzhang-zj
changed the title
|
I believe the bug is in 'self.word_embeddings = transformer_backbone.get_submodule(named_param.replace(".weight", ""))' in peft_model, there don't have 'device', 'self.word_embeddings.device = self.device' may need to be added. |
|
I can reproduce the error. The solution in PEFT would be to change this line: to |
|
@BenjaminBossan Thanks！ |
|
Let's keep this issue open, as we should fix this issue in PEFT :) |
|
This is WIP. I attempted to fix huggingface#1082. While adding tests for the bug, I discovered that I could not make prompt_tuning_init != RANDOM to work. Maybe I'm using it wrong, but I'm not sure what to change.
|
This issue has been automatically marked as stale because it has not had recent activity. If you think this still needs to be addressed please comment on this thread. |
|
On using MPT we have also observed this error. This seems to have been auto-closed due to staleness but wanted to see if this could be reopened since it wasn't resolved. |
|
@evaline-ju There is a PR (#1082) but it's awaiting review. In the meantime, you could manually fix the issue as indicated or install PEFT from the PR branch. |
|
This issue has been automatically marked as stale because it has not had recent activity. If you think this still needs to be addressed please comment on this thread. |
Resolves #1082. Also, adding tests for prompt_tuning_init != RANDOM. --------- Co-authored-by: Mayank Mishra <32954280+mayank31398@users.noreply.github.com>
|
Should be resolved by #1144. |
Resolves huggingface#1082. Also, adding tests for prompt_tuning_init != RANDOM. --------- Co-authored-by: Mayank Mishra <32954280+mayank31398@users.noreply.github.com>
I followed the test process of MPT notebook, but encountered an error during target training: 'Embedding' object has no attribute 'device'. I want to ask how to solve this.
The text was updated successfully, but these errors were encountered: