evals icon indicating copy to clipboard operation
evals copied to clipboard

[evals] Next qwerty letter substitution eval

Open 0xacx opened this issue 1 year ago • 1 comments

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

Next qwerty letter substitution

Eval description

Given a string, this eval asks the model to replace every letter of the string, with the next letter on the right in a QWERTY keyboard. If the next key is a symbol, then it should be replaced with the first letter starting from the left of the same row. Example: "hello" -> "jraap"

What makes this a useful eval?

This eval tests the spatial understanding of the model, as it requires it to create a replacement dictionary based on the position of keys in a QWERTY keyboard. From my limited testing in gpt-3.5-turbo the accuracy is 0%, even for words that were very close to the examples given. (Asking for hell after giving hello as an example, fails!)

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 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

Includes 309 samples.

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 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.

  • [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 granted.

Submit eval

  • [x] I have filled out all required fields in the evals PR form
  • [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": "Given a string, I want you to replace every letter of the string, with the letter that is next to it on the right in a QWERTY keyboard. If the key on the right is a symbol then replace it with the first letter from the left of the same row. So i.e. q should be replaced by w, p should be replaced by q, l should be replaced by a, etc. For example the word hello would become jraap."}, {"role": "user", "content": "abacus"}], "ideal": "snsvid"}
{"input": [{"role": "system", "content": "Given a string, I want you to replace every letter of the string, with the letter that is next to it on the right in a QWERTY keyboard. If the key on the right is a symbol then replace it with the first letter from the left of the same row. So i.e. q should be replaced by w, p should be replaced by q, l should be replaced by a, etc. For example the word hello would become jraap."}, {"role": "user", "content": "apple"}], "ideal": "sqqar"}
{"input": [{"role": "system", "content": "Given a string, I want you to replace every letter of the string, with the letter that is next to it on the right in a QWERTY keyboard. If the key on the right is a symbol then replace it with the first letter from the left of the same row. So i.e. q should be replaced by w, p should be replaced by q, l should be replaced by a, etc. For example the word hello would become jraap."}, {"role": "user", "content": "aardvark"}], "ideal": "sstfbstl"}
{"input": [{"role": "system", "content": "Given a string, I want you to replace every letter of the string, with the letter that is next to it on the right in a QWERTY keyboard. If the key on the right is a symbol then replace it with the first letter from the left of the same row. So i.e. q should be replaced by w, p should be replaced by q, l should be replaced by a, etc. For example the word hello would become jraap."}, {"role": "user", "content": "abbey"}], "ideal": "snnru"}
{"input": [{"role": "system", "content": "Given a string, I want you to replace every letter of the string, with the letter that is next to it on the right in a QWERTY keyboard. If the key on the right is a symbol then replace it with the first letter from the left of the same row. So i.e. q should be replaced by w, p should be replaced by q, l should be replaced by a, etc. For example the word hello would become jraap."}, {"role": "user", "content": "arch"}], "ideal": "stvj"}
{"input": [{"role": "system", "content": "Given a string, I want you to replace every letter of the string, with the letter that is next to it on the right in a QWERTY keyboard. If the key on the right is a symbol then replace it with the first letter from the left of the same row. So i.e. q should be replaced by w, p should be replaced by q, l should be replaced by a, etc. For example the word hello would become jraap."}, {"role": "user", "content": "alligator"}], "ideal": "saaohsypt"}

0xacx avatar Mar 20 '23 14:03 0xacx

Running the eval locally works, so I hope the error was because of the newline in the end of samples.jsonl file.

This is the local run log for just one sample:

$ EVALS_THREADS=1 MAX_SAMPLES=1 oaieval gpt-3.5-turbo next-qwerty-letter-substitution
[2023-03-21 01:03:10,186] [registry.py:145] Loading registry from /Users/achilleas/Workspace/evals/evals/registry/evals
[2023-03-21 01:03:10,221] [registry.py:145] Loading registry from /Users/achilleas/.evals/evals
[2023-03-21 01:03:11,157] [oaieval.py:190] Run started: 2303202303113VVOFNAX
[2023-03-21 01:03:11,159] [data.py:78] Fetching next_qwerty_letter_substitution/samples.jsonl
[2023-03-21 01:03:11,160] [eval.py:30] Evaluating 1 samples
[2023-03-21 01:03:11,170] [eval.py:136] Running in threaded mode with 1 threads!
100%|███████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 1/1 [00:00<00:00,  1.09it/s]
[2023-03-21 01:03:12,095] [record.py:320] Final report: {'accuracy': 0.0}. Logged to /tmp/evallogs/2303202303113VVOFNAX_gpt-3.5-turbo_next-qwerty-letter-substitution.jsonl
[2023-03-21 01:03:12,095] [oaieval.py:221] Final report:
[2023-03-21 01:03:12,095] [oaieval.py:223] accuracy: 0.0
[2023-03-21 01:03:12,096] [record.py:309] Logged 3 rows of events to /tmp/evallogs/2303202303113VVOFNAX_gpt-3.5-turbo_next-qwerty-letter-substitution.jsonl: insert_time=1.176ms

0xacx avatar Mar 20 '23 23:03 0xacx

Thank you for your time in submitting this PR. Character-level reasoning and operations are a well-known failure mode of the model due to a common underlying issue in LLMs. In its current form, this eval does not seem to expose any new gaps in our understanding of model performance. We are increasingly arming our model with tools to offload the computation. Certain tasks that can be easily solved by external tools are less interesting for testing the model's raw capabilities. We also know that this could be solved by giving the model a code interpreter. Model very well knows how to write code to solve such a problem, and that code can be fed to the code interpreter to get the correct answer.

If you're still interested in writing an eval, we've noticed that these criteria make good evals. If you have any particular use case in mind for the model, can you come up with an eval that has some of these attributes?

  • Multi-step reasoning
  • Domain or Application specific
  • Open-Ended responses
  • Complex instructions
  • The eval seems obvious but tricks the model in a novel way

Closing this PR, please open another PR with the provided suggestions.

usama-openai avatar Jun 02 '23 18:06 usama-openai