lmsys

/

vicuna-7b-v1.1

like 75

@weichiang

Sorry for submitting a PR with so much changes.

The tokenizer_config.json content was actually auto generated from

tokenizer  = AutoTokenizer.from_pretrained("lmsys/vicuna-7b-v1.1")
tokenizer.chat_template =  <jinja2 content>
tokenizer.push_to_hub(
                                    repo_id=st.session_state['repo_id'], 
                                    commit_message=commit_message_text_input, 
                                    private=to_private_checkbox, 
                                    token=access_token_no_cache,
                                    create_pr=create_pr_checkbox)

from HuggingSpace.

I think the changes to tokenizer_config.json should be correct as I didn't modify anything other than just adding the chat_template attributes. (Let me know about your thoughts)

As a LLM model user, it is not easy to keep track so many prompt template. This is the UI that I have created to hope that all existing Model Repo could use the new chat_template huggingface feature.