Feature Request: Add Pixtral model support to WebLLM

[mrmehrad]

Dear MLC-AI Team,

First and foremost, I would like to express my sincere appreciation for your outstanding work on WebLLM and the broader MLC ecosystem. Your efforts have significantly advanced the accessibility of large language models in browser environments, and I deeply admire the technical precision and openness of your approach.

I am writing to kindly request support for the Pixtral model within WebLLM. Pixtral has demonstrated strong performance in reasoning tasks and multi-turn dialogue, and its architecture—particularly in distilled variants—appears well-suited for deployment via WebGPU. Given its growing popularity and potential for lightweight inference, I believe Pixtral would be a valuable addition to the MLC model library.

If support for Pixtral is under consideration or already in progress, I would be grateful for any updates. I would also be happy to assist with evaluation, prompt tuning (especially for Persian language use cases), or integration testing if needed.

Thank you again for your dedication to open-source innovation and for making high-performance LLMs more accessible to developers worldwide.

Warm regards,
Mehrad Shekarriz
Developer & Researcher — Persian-centric LLMs and WebGPU Interfaces