Skip to content

add untie_embeddings_and_output_weights config#7

Open
chai-xiaonan wants to merge 1 commit intoflagos-ai:mainfrom
chai-xiaonan:add_embedandoutput_args
Open

add untie_embeddings_and_output_weights config#7
chai-xiaonan wants to merge 1 commit intoflagos-ai:mainfrom
chai-xiaonan:add_embedandoutput_args

Conversation

@chai-xiaonan
Copy link
Copy Markdown

In order to solve the problem that the newly added nemo-bridge function in flagscale fails to obtain the 'untie_embeddings_and_output_weights' parameter configuration when processing the mcore-model, by adding this option, the problem can be permanently resolved. This will enable the acquisition of the parameters of the mcore-model when converting from the hf format to the mcore-model or when converting from the mcore-model to the hf format, and then corresponding processing of the model weights can be carried out.

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

Successfully merging this pull request may close these issues.

1 participant