Add Points-On-Line Eval

[aaronsmithtv]

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 be 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 it 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. Starting April 10, the minimum eval count is 15 samples, we hope this makes it easier to create and contribute evals.

Also, please note that we're using Git LFS for storing the JSON files, so please make sure that you move the JSON file to Git LFS before submitting a PR. Details on how to use Git LFS are available here.

Eval details 📑

Eval name

Points On Line

Eval description

100 sets of vector coordinates in the form of (x, y, z), (x, y, z), with an ideal centre coordinate. The coordinates have a random start position of (-10, -10, -10) to (10, 10, 10) and a furthest maximum distance from origin per-component of 20. All positions are in steps of 0.01 for ease of readability and human understanding.

What makes this a useful eval?

This eval helps gain insight on a GPT model's ability to understand a coordinate space. This is historically a subject that LLMs have been poor in, and provides a reliable, predictable benchmark for whether a model can understand the context of positions within a coordinate space.

gpt-3.5-turbo fails to provide answers that would satisfy the Match class, so I'm now using Include. I've also added some extra complexity, since gpt-4 seemed to do incredibly well on the simpler math with 1 decimal!

Here's the two accuracy reports (0.0 for gpt-3.5-turbo, 0.66 for gpt-4):

[2023-06-03 01:20:18,964] [record.py:341] Final report: {'accuracy': 0.0}. Logged to /tmp/evallogs/230603001824VWSNJZEG_gpt-3.5-turbo_points-on-line.jsonl
[2023-06-03 01:20:18,964] [oaieval.py:147] Final report:
[2023-06-03 01:20:18,964] [oaieval.py:149] accuracy: 0.0

[2023-06-03 01:21:47,663] [record.py:341] Final report: {'accuracy': 0.66}. Logged to /tmp/evallogs/23060300212233RTRLC7_gpt-4_points-on-line.jsonl
[2023-06-03 01:21:47,663] [oaieval.py:147] Final report:
[2023-06-03 01:21:47,663] [oaieval.py:149] accuracy: 0.66

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:

• [x] 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.
• [x] Contains failures where a human can do the task, but either GPT-4 or GPT-3.5-Turbo could not.
• [x] 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.
• [x] Include at least 15 high-quality examples.

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

Unique eval value

These evals come with a generator script that can create new coordinate datasets very quickly. It can also be expanded to account for future, more difficult scopes of this test, such as larger distances, greater floating point deviation, and total numbers of points to calculate in a space.

Eval structure 🏗️

Your eval should

• [x] Check that your data is in evals/registry/data/{name}
• [x] Check that your YAML is registered at evals/registry/evals/{name}.yaml
• [x] 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).

• [x] 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.

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

Limited availability acknowledgment

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 the 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.

• [x] 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 be granted.

Submit eval

• [x] I have filled out all required fields of this form
• [x] I have used Git LFS for the Eval JSON data
• [x] (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": "You will be provided with the end points of a line in 3 dimensions. Please calculate and return only the midpoint of this line, in this format: (x, y, z)"}, {"role": "user", "content": "(4.10, -4.98, -6.99), (14.28, -23.12, 4.89)"}], "ideal": "(9.19, -14.05, -1.05)"}
{"input": [{"role": "system", "content": "You will be provided with the end points of a line in 3 dimensions. Please calculate and return only the midpoint of this line, in this format: (x, y, z)"}, {"role": "user", "content": "(-1.98, -5.97, -9.94), (-21.94, -19.87, 2.02)"}], "ideal": "(-11.96, -12.92, -3.96)"}
{"input": [{"role": "system", "content": "You will be provided with the end points of a line in 3 dimensions. Please calculate and return only the midpoint of this line, in this format: (x, y, z)"}, {"role": "user", "content": "(2.09, 9.92, 1.06), (4.13, 27.90, -5.14)"}], "ideal": "(3.11, 18.91, -2.04)"}
{"input": [{"role": "system", "content": "You will be provided with the end points of a line in 3 dimensions. Please calculate and return only the midpoint of this line, in this format: (x, y, z)"}, {"role": "user", "content": "(7.07, -1.05, 0.94), (-13.07, -11.17, 17.10)"}], "ideal": "(-3.00, -6.11, 9.02)"}
{"input": [{"role": "system", "content": "You will be provided with the end points of a line in 3 dimensions. Please calculate and return only the midpoint of this line, in this format: (x, y, z)"}, {"role": "user", "content": "(6.90, 4.92, 1.93), (0.74, -11.14, -4.11)"}], "ideal": "(3.82, -3.11, -1.09)"}

[aaronsmithtv]

Hey @joe-at-openai and @usama-openai , many thanks for the helpful suggestions in improving my eval in #288 ! I've taken a fresh look at the prompt and results, and I've added a new set of samples for you to take a look at. Please let me know if there is anything else I can do to improve this time around, as the feedback is greatly appreciated.

For a quick overview of the differences, the point stepping is now set at 0.01 and not 0.1 as gpt-4 excels at the 0.1 evals. The evals now use Include and not Match, and have a shorter, clearer prompt. I can include the generation script if needed! :)

[aaronsmithtv]

Many thanks for reviewing my PR @usama-openai ! I've just added the requested changes alongside the generator script. Please let me know if there are any extra changes you'd like me to take a look at implementing :)

[aaronsmithtv]

My apologies! I noticed it as the tests were coming through. That should be fixed now :)

[usama-openai]

You should see GPT-4 API access enabled in your account in the next few days.

[aaronsmithtv]

Great, many thanks @andrew-openai and @usama-openai ! I am happy for my GPT-4 access to go to someone else, I already have it :)