Nouha DziriandGitHub 5eefbacd97 Generate and score responses (LLM-as-judge) with API-based models (#252) ...

* move rejection sampling within its own directory

* move rejection sampling within its own directory

* add api-based models to generating completions

* debug

* debug

* debug

* debug

* debug

* do not apply tokenization with chat template

* change the tokenizer

* change the tokenizer

* fix template

* fix arg issue

* add mode to arg

* add mode to arg

* change the skill

* remove main from api_generate.py

* modify template chat

* modify template chat

* return only text

* return only text

* return only text

* fix dict

* fix dict

* fix dict

* fix dict

* fix dict

* fix dict

* add LM as a scorer

* add LM as a scorer

* add LM as a scorer

* add LM as a scorer

* change the name of the model

* fix rank bug

* process score

* process score

* edit the template

* extract score

* compute scores for reference responses

* compute scores for reference responses

* add tqdm

* add tqdm

* fix score bug

* fix item bug

* fix item bug

* fixed code style

* fixed code style

* .idea added to .gitignore

* .idea dir removed

* remove unused llm_as_judge file

* Update rejection_sampling.md

* fix the path to rejection sampling files

* update readme

* remove unused args

* use args

* remove args

* fix chatgpt completion

* fix chatgpt completion

* fix chatgpt completion

* fix bug

* fix format conversation

* fix minor bug

* fix minor bug

* fix minor bug

* fix minor bug

* fix parsing response

* test judgment

* update format

* update format

* debug

* debug

* debug

* fix

* format fixed

* format fixed

* fix bug

* fix bug

* fix bug

* fix bug

* add comment

2024-08-14 11:37:10 -04:00

0 lines

0 B

Python

Raw Permalink Blame History

The file is empty.