Supported vLLM features

[Sun-Xiaohui]

Hello, I am interested in MooreThreads GPUs. I want to know what vLLM features are supported on MUSA? For example, If I want to deploy Dflash on MUSA, What tasks do I need to do and how difficult will they be?

[yeahdongcn]

Thanks for reaching out! We are currently working on performance improvements (v0.20.0-dev).

As for DFlash, since it is a relatively new feature in vLLM, we plan to investigate it in the next vLLM upgrade cycle (e.g. v0.22.0). At the moment, we support and have tested Eagle3 as the speculative decoding method.

[Sun-Xiaohui]

Thanks for your reply. Where can I view the list of currently supported models and features?

[yeahdongcn]

Thanks for your reply. Where can I view the list of currently supported models and features?

@zichen-kuuga @Joey-gvwal Since we are doing this internally, could we create several new documents to track these items? Feel free to open new PRs for them. Thanks!