Beidi Chen

@beidichen

Asst. Prof @CarnegieMellon, @amazon Scholar, Prev: Visiting Researcher @Meta, Postdoc @Stanford, Ph.D. @RiceUniversity, Large-Scale ML, a fan of Dota2.

following
followers

lithos-metal is being open-sourced, with a claimed peak of 200-plus tokens per second per user on one Apple M5 Max

The developer says megakernels and DSpark speculative decoding run Qwen3.8-27B at that peak, and shares the code.

3 Sources, 12d ago, first seen 12d ago

ServeLearnBench tests continual self-improvement in AI agents

A post introducing the benchmark calls agents that get smarter as people use them “the next frontier.”

1 Source, 13d ago, first seen 13d ago

Inco AI Releases DFlash 2 for Faster Qwen Inference

Researchers react to the speculative decoding update for Qwen models on Apple hardware.

10 Sources, 63d ago, first seen 63d ago

Featured in

lithos-metal is being open-sourced, with a claimed peak of 200-plus tokens per second per user on one Apple M5 Max

ServeLearnBench tests continual self-improvement in AI agents

Inco AI Releases DFlash 2 for Faster Qwen Inference

Featured in

lithos-metal is being open-sourced, with a claimed peak of 200-plus tokens per second per user on one Apple M5 Max

ServeLearnBench tests continual self-improvement in AI agents

Inco AI Releases DFlash 2 for Faster Qwen Inference