Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

[Bugfix] Ensure special tokens are properly filtered out for guided structured output with MistralTokenizer #10363

Conversation

gcalmettes
Copy link
Contributor

@gcalmettes gcalmettes commented Nov 15, 2024

This PR improves the support of the vllm MistralTokenizer with the guided structured output functionalities.

In particular:

  • the version of lm-format-enforcer is bumped to the latest version, so it includes the support of the vllm MistralTokenizer (feat: add support for vLLM Mistral Tokenizer noamgat/lm-format-enforcer#142)
  • the vllm MistralTokenizer methods leveraged by the structured output libraries are now properly populated so that the list of possible tokens generated based on the chosen grammar/constraints (e.g.: json output) correctly filters out the special tokens. Note that this problem was present before, but masked by the fact that the special tokens were automatically filtered out of the conversion from ids to string: the recent vllm support for automatic mistral tool parsing allowed to reveal the bug (also see this comment #10333). As a benefit, this should also make the structured output generation by Mistral models more reliable.

cc: @patrickvonplaten

Example: code that would break without this PR but passes with it

"""
vllm openai server started with the following arguments:
    --model=mistralai/Pixtral-12B-2409
    --guided-decoding-backend=lm-format-enforcer 
    --enable-auto-tool-choice 
    --tool-call-parser=mistral 
    --tokenizer-mode=mistral
"""

from openai import OpenAI
import json
from pydantic import BaseModel

client = OpenAI(
    base_url="http://localhost:8000/v1",
    api_key="none",
)

######################################
# Example 1: using `client.beta.chat.completions.parse`
######################################

class CalendarEvent(BaseModel):
    name: str
    date: str
    participants: list[str]

completion = client.beta.chat.completions.parse(
    model="mistralai/Pixtral-12B-2409",
    messages=[
        {"role": "system", "content": "Extract the event information."},
        {"role": "user", "content": "Alice and Bob are going to a science fair on Friday."},
    ],
    response_format=CalendarEvent,
    max_completion_tokens=10000,
)

event = completion.choices[0].message.parsed
print(event.__dict__)

######################################
# Example 2: using `client.chat.completions.create`
######################################

class Step(BaseModel):
    explanation: str
    output: str

class MathReasoning(BaseModel):
    steps: list[Step]
    final_answer: str

completion = client.chat.completions.create(
    model="mistralai/Pixtral-12B-2409",
    messages=[
        {"role": "system", "content": "You are a helpful math tutor. Guide the user through the solution step by step."},
        {"role": "user", "content": "how can I solve 8x + 7 = -23"}
    ],
    response_format={
        "type": "json_schema",
        "json_schema": {
            "name": "toto",
            "schema": MathReasoning.model_json_schema()
        }
    },
    max_completion_tokens=10000,
)

math_reasoning = completion.choices[0].message.content
print(json.loads(json.dumps(math_reasoning, indent=2)))

Copy link

👋 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 starts running only a small and essential subset of CI tests to quickly catch errors. You can run other CI tests on top of those by going to your fastcheck build on Buildkite UI (linked in the PR checks section) and unblock them. If you do not have permission to unblock, ping simon-mo or khluu to add you in our Buildkite org.

Once the PR is approved and ready to go, your PR reviewer(s) can run CI to test the changes comprehensively before merging.

To run CI, PR reviewers can do one of these:

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

🚀

@mergify mergify bot added the ci/build label Nov 15, 2024
@gcalmettes gcalmettes changed the title [Bugfix] Ensure special tokens are properly filtered out for guided structured output [Bugfix] Ensure special tokens are properly filtered out for guided structured output with MistralTokenizer Nov 15, 2024
@gcalmettes
Copy link
Contributor Author

@DarkLight1337 it's a follow up of #10333

@DarkLight1337
Copy link
Member

cc @simon-mo

@DarkLight1337 DarkLight1337 added the ready ONLY add when PR is ready to merge/full CI is needed label Nov 15, 2024
@gcalmettes
Copy link
Contributor Author

@DarkLight1337 unfortunately, we did not get lucky and got timed-out on the Async Engine, Inputs, Utils, Worker Test test. Any chance you restart this one ? 🙏

@DarkLight1337 DarkLight1337 enabled auto-merge (squash) November 15, 2024 14:02
@DarkLight1337 DarkLight1337 merged commit 691a3ec into vllm-project:main Nov 15, 2024
78 of 81 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
ci/build ready ONLY add when PR is ready to merge/full CI is needed
Projects
None yet
Development

Successfully merging this pull request may close these issues.

2 participants