[eval] Chinese Idioms evulation

[robinzixuan]

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

chinese_idioms

Eval description

Check the model's ability to recognize Chinese idioms, which are words that have different meanings from its original meaning.

What makes this a useful eval?

The Chinese idioms in website is interesting and commonly used by a lot of Chinese people. However, the GPT4 and GPT3.5 fail to explain the meaning of the idioms with a correct explanation.

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

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

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 commits on 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
• (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": "user", "content": "请解释下面词语的意思,请使用英文回答。\n---\n伟光正"}], "ideal": ["From the idiomatic phrase 'the great, glorious and correct Chinese Communist Party', it can also refer to a person associated with the Chinese Communist Party."]}
{"input": [{"role": "user", "content": "请解释下面词语的意思,请使用英文回答。\n---\n赵家人"}], "ideal": ["From Lu Xun's famous middle-grade novel 'A Q Zhengzhuan', it generally refers to the powerful and noble class of the Chinese Communist Party. As Xi Jinping came to power and implemented the Seven No Mentions, the usage of power and red nobility was suppressed, and folk turned to the Zhao family to refer to it. Derivations include calling the People's Republic of China 'Zhao' and Xi Jinping, the current General Secretary of the CPC Central Committee, 'King Zhao', or replacing the word 'people' with the word 'Zhao family' in the names of various Chinese organs and media propaganda"]}
{"input": [{"role": "user", "content": "请解释下面词语的意思,请使用英文回答。\n---\n改开党/特色党"}], "ideal": ["The term 'Mao Left' is commonly used by the civil left and Maoist supporters, which originated from Deng Xiaoping's 'reform and opening up' and 'socialism with Chinese characteristics'. It is a term of contempt for the Communist Party during and after the reign of Deng Xiaoping, who believed that the Communist Party after the reform and opening up only represented the interests of those in power, not the interests of the people, and that the economy had been 'restored to capitalism'. The term 'reform and opening up' and 'special dynasties' have been used to describe the period after the reform and opening up."]}
{"input": [{"role": "user", "content": "请解释下面词语的意思,请使用英文回答。\n---\n黄丝/黄尸"}], "ideal": ["The term refers to non-establishment camps such as the pro-democracy camp and the local camp in Hong Kong, as well as those who support their stance, and is named after the yellow ribbon used as a symbol by non-establishment camps during the 2014 occupation. Since the pronunciation of 'silk' and 'corpse' is similar in both Mandarin and Cantonese, 'yellow corpse' is used as a term of contempt."]}
{"input": [{"role": "user", "content": "请解释下面词语的意思,请使用英文回答。\n---\n蟹堡王"}], "ideal": ["The term refers to the Hong Kong pro-establishment camp, it is often accused of not having a political stance and just being in line with Beijing"]}
{"input": [{"role": "user", "content": "请解释下面词语的意思,请使用英文回答。\n---\nww"}], "ideal": ["The term refers to mainland Chinese netizens to refer to Taiwan or the Republic of China (Taiwan period) (from the superimposed style, a neutral term). In January 2022, Taiwan Affairs Office spokesperson Zhu Fenglian said that the word Wanwan is a nickname for the Taiwanese people 'Mengmeng' by the Chinese mainlanders"]}

[robinzixuan]

I make another evaluation of the reverse shell detection. I looks like also shown here. should I revert that

[robinzixuan]

I revert it first

[usama-openai]

Thanks for the contribution. I would like to request a change.

The evaluation method being used to evaluate the answer is Match which only works when an exact match is found. The user needs to provide very specific instructions to the model about the output format, and the ideal answer should be in that expected format. In this dataset, the model is being asked to translate the provided Chinese idiom into English. The input prompt and the ideal answer are not formatted in a way to be matched word for word. For example, one ideal answer is: The term refers to people who indulge in online shopping. Also known as 'chopper', 'chopper gang', etc. The model may provide an answer like: "Duo Shou Dang" (剁手党) is a Chinese slang term that refers to people who are addicted to online shopping and spend a lot of money on it, often to the point of financial difficulty. In English, it can be translated as "chop hand party" or "shopaholic".. which is also correct, but the Match evaluation method will mark it false.

I would recommend you either make it a multiple-choice question and ask the model to choose the correct translation from the provided options or use the ModelGraded evaluation method here. You can read the documentation or look at the other evals to get a good idea about the ModelGraded method.

We would love to review the PR again after the suggested changes.

[robinzixuan]

I make the change to the ModelGraded, but I am not sure whether it is ok. Thanks

[usama-openai]

This PR looks in good shape now. I'm approving this PR.

[usama-openai]

For this eval, GPT-3.5 doesn't perform well for grading. So, GPT-4 should be used as the grader model.

[robinzixuan]

Thanks

[robinzixuan]

I would really appreciate access to the 8k and 32k GPT-4 APIs, and am happy to do more evals if that helps!