Successful reproduction of the experiments on APPS by pure GPT3.5 #9
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
fix: set num_completions and remove header. fix: details of prompts. fix: prompts. config: modify .gitignor fix: adjust prompts. fix: max_tokens. fix: no_tests. fix: no_tests -> add asserts. fix: finish auto gen tests. fix: no_tests. fix: allow test failed, modify codeT score part. fix: prompts. fix: prompts. fix: adjust prompts. feat: Add logit_bias to force gpt use implemented functions. fix: delete useless code. fix: logit_bias -> prompts. fix: prompts. feat: Add num_completions and save_path args. feat: add __init__.py, make parsel like a package. fix: generate_tests = True. fix: found_successful_generation. fix: grammar mistake in prompts. fix: timeout exception, use single process. fix: code transform to remove implemented functions. fix: prompts. fix: restore logit_bias. fix: Allow nested functions overwrite implemented func fix: sleep. fix: remove generate_tests. fix: adjust prompts. fix: adjust prompts. fix: timeout. fix: num_completions. fix: product sample. fix: seed before shuffle. fix: multiprocess. fix: timeout. feat: handle MLE. fix: MemoryError.
|
Hey Yutong, could you share the modifications related to evaluations as well? I'm trying to reproduce the results on apps (27/100) according to your post. |
Sorry, since a long time passed, I forgot many details about evaluations. See https://github.com/wyt2000/Automatic-ANPL/tree/apps for help. |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Since Codex was deprecated by OpenAI, I tried to reproduce the experiments on the dataset APPS in Parsel paper by pure GPT3.5. Thanks to the code in branch
saycan, I fully understood your evalutation method. After a tough struggling to modify the prompts and Parsel itself, I finally reproduced a part of experiments mentioned in chapter 3.1 of the paper and even got better results: the pure GPT-3.5 version parsel(8x16) solved 27 of 100 randomly sampled competition-level problem in APPS. I offer the modified code for someone to use in the future.