[Frontend] Represent tokens with identifiable strings

[ezliu]

Many model tokens are not representable as ASCII strings. For example, emojis (e.g., 🤣) are represented as 2 tokens in the Llama3 tokenizer, where the first token represents the first 3 Unicode bytes of the emoji, and the second token represents the last byte. In the current implementation of the OpenAI-compatible server, the first token is represented as an empty string, so it is not possible to determine what the actual token ID is.

Here is a concrete example. If we spin up a Llama3 8B server as follows:

python -m vllm.entrypoints.openai.api_server --model meta-llama/Meta-Llama-3-8B --tensor-parallel-size 8 --max-logprobs 20

Then, we hit it with a request including log probs:

curl http://localhost:8000/v1/completions \
    -H “Content-Type: application/json” \
    -d ‘{                                           
        “model”: “meta-llama/Meta-Llama-3-8B”,
        “prompt”: “😊😄😃😁😎”,
        “max_tokens”: 20,                           
        “temperature”: 1,                           
        “top_p”: 0.95,                              
        “n”: 1,                                     
        “echo”: true,                             
        “logprobs”: 2                               
    }’

The "tokens" field of the output that I receive is:

        “tokens”: [                                                                                       
          “token_id:128000“, “”, “”, “”, “”, “”, “”, “”, “”, “”, “”, “”, “”, “🤩”, “”, “😨”, “”, “”, “🤣”, “”, “”, ” 🤪”, “”, “💣”, “”, “”, “👍”, “”, “”, “👉”, “”
        ],

All of the empty strings correspond to partial Unicode code points that are not being properly represented.

This PR fixes this behavior by adding a --return-tokens-as-token-ids flag to the api_server.py. When this flag is passed, the tokens are represented as "token_id:{token_id}", similar to how the OpenAI server represents these tokens as "bytes:{byte_value_corresponding_to_the_token}". We choose to directly return the token IDs, since HuggingFace tokenizers are unable to map byte values to the corresponding tokens.

As a concrete example, the above curl request when the --return-tokens-as-token-ids is passed to the server spin up yields this "tokens" field instead:

"tokens":["<|begin_of_text|>","token_id:76460","token_id:232","token_id:76460","token_id:226","token_id:76460","token_id:225","token_id:76460","token_id:223","token_id:76460","token_id:236","token_id:76460","token_id:227","token_id:76460","token_id:232","token_id:76460","token_id:224","token_id:9468","token_id:97","token_id:96","token_id:76460","token_id:225","token_id:76460","token_id:224","token_id:76460","token_id:227","token_id:76460","token_id:228","token_id:76460","token_id:224","token_id:76460"]

The default behavior of the server is unchanged.

BEFORE SUBMITTING, PLEASE READ THE CHECKLIST BELOW AND FILL IN THE DESCRIPTION ABOVE

---

PR Checklist (Click to Expand)

Thank you for your contribution to vLLM! Before submitting the pull request, please ensure the PR meets the following criteria. This helps vLLM maintain the code quality and improve the efficiency of the review process.

PR Title and Classification

Only specific types of PRs will be reviewed. The PR title is prefixed appropriately to indicate the type of change. Please use one of the following:

• [Bugfix] for bug fixes.
• [CI/Build] for build or continuous integration improvements.
• [Doc] for documentation fixes and improvements.
• [Model] for adding a new model or improving an existing model. Model name should appear in the title.
• [Frontend] For changes on the vLLM frontend (e.g., OpenAI API server, LLM class, etc.)
• [Kernel] for changes affecting CUDA kernels or other compute kernels.
• [Core] for changes in the core vLLM logic (e.g., LLMEngine, AsyncLLMEngine, Scheduler, etc.)
• [Hardware][Vendor] for hardware-specific changes. Vendor name should appear in the prefix (e.g., [Hardware][AMD]).
• [Misc] for PRs that do not fit the above categories. Please use this sparingly.

Note: If the PR spans more than one category, please include all relevant prefixes.

Code Quality

The PR need to meet the following code quality standards:

• We adhere to Google Python style guide and Google C++ style guide.
• Pass all linter checks. Please use format.sh to format your code.
• The code need to be well-documented to ensure future contributors can easily understand the code.
• Include sufficient tests to ensure the project to stay correct and robust. This includes both unit tests and integration tests.
• Please add documentation to docs/source/ if the PR modifies the user-facing behaviors of vLLM. It helps vLLM user understand and utilize the new features or changes.

Notes for Large Changes

Please keep the changes as concise as possible. For major architectural changes (>500 LOC excluding kernel/data/config/test), we would expect a GitHub issue (RFC) discussing the technical design and justification. Otherwise, we will tag it with rfc-required and might not go through the PR.

What to Expect for the Reviews

The goal of the vLLM team is to be a transparent reviewing machine. We would like to make the review process transparent and efficient and make sure no contributor feel confused or frustrated. However, the vLLM team is small, so we need to prioritize some PRs over others. Here is what you can expect from the review process:

• After the PR is submitted, the PR will be assigned to a reviewer. Every reviewer will pick up the PRs based on their expertise and availability.
• After the PR is assigned, the reviewer will provide status update every 2-3 days. If the PR is not reviewed within 7 days, please feel free to ping the reviewer or the vLLM team.
• After the review, the reviewer will put an action-required label on the PR if there are changes required. The contributor should address the comments and ping the reviewer to re-review the PR.
• Please respond to all comments within a reasonable time frame. If a comment isn't clear or you disagree with a suggestion, feel free to ask for clarification or discuss the suggestion.

Thank You

Finally, thank you for taking the time to read these guidelines and for your interest in contributing to vLLM. Your contributions make vLLM a great tool for everyone!

[github-actions]

👋 Hi! Thank you for contributing to the vLLM project.
Just a reminder: PRs would not trigger full CI run by default. Instead, it would only run fastcheck CI which consists a small and essential subset of CI tests to quickly catch errors. You can run other CI tests on top of default ones by unblocking the steps in your fast-check build on Buildkite UI.

Once the PR is approved and ready to go, please make sure to run full CI as it is required to merge (or just use auto-merge).

To run full CI, you can do one of these:

• Comment /ready on the PR
• Add ready label to the PR
• Enable auto-merge.

🚀

[simon-mo]

Thank you for submitting the PR! I'm curious what would OpenAI return in this case?

[ezliu]

OpenAI returns the tokens under a similar "bytes:" prefix. For example, here is the "tokens" field from a similar request from GPT-3.5-Turbo with emojis.

tokens=['bytes:\\xf0\\x9f\\x98', 'bytes:\\x8d', 'bytes:\\xf0\\x9f\\x98', 'bytes:\\x98', 'bytes:\\xf0\\x9f\\x98', 'bytes:\\x97', 'bytes:\\xf0\\x9f\\x98', 'bytes:\\x99', 'bytes:\\xf0\\x9f\\x98', 'bytes:\\x9a', 'bytes:\\xf0\\x9f\\x98', 'bytes:\\x9c', 'bytes:\\xf0\\x9f\\x98', 'bytes:\\x9d', 'bytes:\\xf0\\x9f\\x98', 'bytes:\\x9b', 'bytes:\\xf0\\x9f\\x98', 'bytes:\\xb3', 'bytes:\\xf0\\x9f\\x98', 'bytes:\\x81']

I'm proposing a new prefix "token_id", since there is not a good way to convert these byte representations back into tokens via HuggingFace tokenizers. This property isn't needed from the OpenAI interface, because nobody needs to convert OpenAI model tokens into token IDs, whereas people may want to do this for an open-weight model.

[DarkLight1337]

This is a good idea! Could you update the code to resolve the merge conflicts?

It would also be great if you could add tests for this new behaviour.

[ezliu]

Thanks for the comments! Merged and added tests.