Add eval for unified patch diffs

[AndreBaltazar8]

Thank you for contributing an eval! ♥️

🚨 Please make sure your PR follows these guidelines, failure to follow the guidelines below will result in the PR being closed automatically. Note that even if the criteria are met, that does not guarantee the PR will be merged nor GPT-4 access granted. 🚨

PLEASE READ THIS:

In order for a PR to be merged, it must fail on GPT-4. We are aware that right now, users do not have access, so you will not be able to tell if the eval fails or not. Please run your eval with GPT-3.5-Turbo, but keep in mind as we run the eval, if GPT-4 gets higher than 90% on the eval, we will likely reject since GPT-4 is already capable of completing the task.

We plan to roll out a way for users submitting evals to see the eval performance on GPT-4 soon. Stay tuned! Until then, you will not be able to see the eval performance on GPT-4. We encourage partial PR's with ~5-10 example that we can then run the evals on and share the results with you so you know how your eval does with GPT-4 before writing all 100 examples.

Eval details 📑

Eval name

Unified Patch Diffs

Eval description

Make a correct Unified Patch from the input file and a malformed patch

What makes this a useful eval?

The model seems to be bad at differentiating lines and counts. This example has both at the same time, it needs to know how many lines were in the file and how many got added or removed.

Criteria for a good eval ✅

Below are some of the criteria we look for in a good eval. In general, we are seeking cases where the model does not do a good job despite being capable of generating a good response (note that there are some things large language models cannot do, so those would not make good evals).

Your eval should be:

• Thematically consistent: The eval should be thematically consistent. We'd like to see a number of prompts all demonstrating some particular failure mode. For example, we can create an eval on cases where the model fails to reason about the physical world.
• Contains failures where a human can do the task, but either GPT-4 or GPT-3.5-Turbo could not.
• Includes good signal around what is the right behavior. This means either a correct answer for Basic evals or the Fact Model-graded eval, or an exhaustive rubric for evaluating answers for the Criteria Model-graded eval.
• Include at least 100 high quality examples (it is okay to only contribute 5-10 meaningful examples and have us test them with GPT-4 before adding all 100)

If there is anything else that makes your eval worth including, please document it below.

Unique eval value

Insert what makes your eval high quality that was not mentioned above. (Not required)

This seems invaluable for it to know how to generate proper patches of code, it needs to know how to count lines so it can produce correct patches that could be applied to code. Just knowing how to count lines in general is a very good skill for other aspects as well.

Eval structure 🏗️

Your eval should

• Check that your data is in evals/registry/data/{name}
• Check that your yaml is registered at evals/registry/evals/{name}.yaml
• Ensure you have the right to use the data you submit via this eval

(For now, we will only be approving evals that use one of the existing eval classes. You may still write custom eval classes for your own cases, and we may consider merging them in the future.)

Final checklist 👀

Submission agreement

By contributing to Evals, you are agreeing to make your evaluation logic and data under the same MIT license as this repository. You must have adequate rights to upload any data used in an Eval. OpenAI reserves the right to use this data in future service improvements to our product. Contributions to OpenAI Evals will be subject to our usual Usage Policies (https://platform.openai.com/docs/usage-policies).

• I agree that my submission will be made available under an MIT license and complies with OpenAI's usage policies.

Email address validation

If your submission is accepted, we will be granting GPT-4 access to a limited number of contributors. Access will be given to the email address associated with the merged pull request.

• I acknowledge that GPT-4 access will only be granted, if applicable, to the email address used for my merged pull request.

Limited availability acknowledgement

We know that you might be excited to contribute to OpenAI's mission, help improve our models, and gain access to GPT-4. However, due to the requirements mentioned above and high volume of submissions, we will not be able to accept all submissions and thus not grant everyone who opens a PR GPT-4 access. We know this is disappointing, but we hope to set the right expectation before you open this PR.

• I understand that opening a PR, even if it meets the requirements above, does not guarantee the PR will be merged nor GPT-4 access granted.

Submit eval

• I have filled out all required fields in the evals PR form
• (Ignore if not submitting code) I have run pip install pre-commit; pre-commit install and have verified that black, isort, and autoflake are running when I commit and push

Failure to fill out all required fields will result in the PR being closed.

Eval JSON data

Since we are using Git LFS, we are asking eval submitters to add in as many Eval Samples (at least 5) from their contribution here:

View evals in JSON

Eval

{"input": [{"role": "system", "content": "Task: Output the correct patch from the given input and malformed patch with incorrect line numbers in Unified Diff format."}, {"role": "user", "content": "Given this input file ```\n701455\n753947\n952545\n694658\n724002\n158388\n379555\n997602\n833233\n689750\n417479\n468027\n69993\n```\n and this malformed patch ```\n--- file_58566.txt\n+++ file_58566.txt\n@@ X,X X,X @@\n 701455\n 753947\n 952545\n-694658\n 724002\n 158388\n-379555\n 997602\n 833233\n 689750\n``` output just the corrected patch, nothing else."}], "ideal": ["--- file_58566.txt\n+++ file_58566.txt\n@@ -1,10 +1,8 @@\n 701455\n 753947\n 952545\n-694658\n 724002\n 158388\n-379555\n 997602\n 833233\n 689750\n"]}
{"input": [{"role": "system", "content": "Task: Output the correct patch from the given input and malformed patch with incorrect line numbers in Unified Diff format."}, {"role": "user", "content": "Given this input file ```\n137132\n958115\n942829\n971026\n123400\n706739\n231446\n112618\n75014\n250593\n862412\n657169\n```\n and this malformed patch ```\n--- file_697676.txt\n+++ file_697676.txt\n@@ X,X X,X @@\n 958115\n 942829\n 971026\n-123400\n 706739\n 231446\n 112618\n 75014\n-250593\n 862412\n 657169\n``` output just the corrected patch, nothing else."}], "ideal": ["--- file_697676.txt\n+++ file_697676.txt\n@@ -2,11 +2,9 @@\n 958115\n 942829\n 971026\n-123400\n 706739\n 231446\n 112618\n 75014\n-250593\n 862412\n 657169\n"]}
{"input": [{"role": "system", "content": "Task: Output the correct patch from the given input and malformed patch with incorrect line numbers in Unified Diff format."}, {"role": "user", "content": "Given this input file ```\n550481\n5805\n420810\n891010\n385616\n444523\n886228\n451405\n979223\n105265\n770968\n830129\n```\n and this malformed patch ```\n--- file_891507.txt\n+++ file_891507.txt\n@@ X,X X,X @@\n 550481\n-5805\n 420810\n-891010\n 385616\n 444523\n 886228\n``` output just the corrected patch, nothing else."}], "ideal": ["--- file_891507.txt\n+++ file_891507.txt\n@@ -1,7 +1,5 @@\n 550481\n-5805\n 420810\n-891010\n 385616\n 444523\n 886228\n"]}
{"input": [{"role": "system", "content": "Task: Output the correct patch from the given input and malformed patch with incorrect line numbers in Unified Diff format."}, {"role": "user", "content": "Given this input file ```\n625496\n869955\n623320\n136931\n697635\n327957\n310820\n317489\n174338\n763637\n28132\n991157\n956340\n901946\n93178\n929093\n503735\n756683\n600214\n880008\n```\n and this malformed patch ```\n--- file_604910.txt\n+++ file_604910.txt\n@@ X,X X,X @@\n 869955\n 623320\n 136931\n-697635\n 327957\n 310820\n 317489\n 174338\n 763637\n 28132\n-991157\n 956340\n 901946\n 93178\n``` output just the corrected patch, nothing else."}], "ideal": ["--- file_604910.txt\n+++ file_604910.txt\n@@ -2,14 +2,12 @@\n 869955\n 623320\n 136931\n-697635\n 327957\n 310820\n 317489\n 174338\n 763637\n 28132\n-991157\n 956340\n 901946\n 93178\n"]}
{"input": [{"role": "system", "content": "Task: Output the correct patch from the given input and malformed patch with incorrect line numbers in Unified Diff format."}, {"role": "user", "content": "Given this input file ```\n13777\n470864\n706722\n268961\n981555\n636859\n109180\n923125\n659453\n110649\n859417\n330554\n401691\n13351\n```\n and this malformed patch ```\n--- file_610459.txt\n+++ file_610459.txt\n@@ X,X X,X @@\n 923125\n 659453\n 110649\n-859417\n 330554\n-401691\n 13351\n``` output just the corrected patch, nothing else."}], "ideal": ["--- file_610459.txt\n+++ file_610459.txt\n@@ -8,7 +8,5 @@\n 923125\n 659453\n 110649\n-859417\n 330554\n-401691\n 13351\n"]}

[AndreBaltazar8]

Note on the eval creation:

These patches were generated automatically from a python script that generates random numbers on a file. Random lines from this file are removed, and then a patch was generated to be put in "ideal", line counts were then replaced by X to be used in the prompt.

I think this is valuable with the randomness, if it would be better to have code patches or other different contents let me know!

[unicomp21]

@andrew-openai, for gpt-3.5-turbo and gpt-4, if unified diff format is not reliable, won't we have a bunch of angry devs/coders? Is this sort of a base requirement? copy/paste can only go so far?

or perhaps there is some other length delimited format, perhaps custom, we should be using for diffs?

[unicomp21]

normal workflow. apply diff, lint, run tests, repeat. should we mark this issue as critical @andrew-openai ?

[unicomp21]

#588

[andrew-openai]

Thanks for the eval, looks interesting! Merging (will add gpt4 access)

[AndreBaltazar8]

Thank you @andrew-openai. I have just realized I don't use the email I created the PR from as my Open AI account, but I've created the account under that email now. Sorry for the inconvenience.

[unicomp21]

@andrew-openai when it comes to coding using gpt-4, do we think it's possible to get a reliable unidiff? Or is another approach needed?

[AndreBaltazar8]

Got the access 👍 Thank you :)

[unicomp21]

@AndreBaltazar8 I'm trying to do an eval for length delimited strings, but am having trouble understanding the jsonl file?