Less Wright

icon indicating a website link https://www.linkedin.com/in/lessw2020 icon indicating an email [email protected]

icon company Cursor icon location Seattle, WA USA icon location AI Software Engineer - Cursor, PyTorch Partner Eng, Lead - Meta, Principal Software Engineer - Audere Software Architect - X10 Wireless Dev/PM - Microsoft

Results 97 comments of Less Wright

Possible to integrate DeepEP?

Sep issue for DeepGEMM here: https://github.com/pytorch/torchtitan/issues/889

[DeepSeek MoE] current workstream planning

#2 = PR https://github.com/pytorch/torchtitan/pull/1127

[Feature request] Integrate DeepGEMM

Well the gemms are very performant but these are inference only. They didn't release the backward portion ala wgrad. From their issues discussion it seems they are considering releasing but...

[Feature request] Integrate DeepGEMM

working on triton implementation to support both inference and training. Bf16 version forward in testing now.

[Feature request] Integrate DeepGEMM

Progress update: We have landed a forward MG * NG group Gemm for deepseek inference this week (bf16)...you can run it using generate.py. This also has backward kernels but needs...

[Feature request] Integrate DeepGEMM

Thanks @vwxyzjn for the update! We have a cleaner version of deepseek now, so we can potentially integrate there, or just jump to mxfp8 directly.

[Feature request] Integrate DeepGEMM

hi @ajWithNucleus, I'm no longer working on Titan but maybe @tianyu-l or @danielvegamyhre can provide an update if any plans to integrate these training kernels.