zhexiang zhang's picture

1 1

zhexiang zhang

Somniloquy

AI & ML interests

None yet

Recent Activity

upvoted a paper 10 days ago

Janus: Disaggregating Attention and Experts for Scalable MoE Inference

new activity over 1 year ago

mistralai/Mistral-7B-Instruct-v0.2:deepspeed inference tensor parallelism memory footprint doesn't decrease with deepspeed tp_size increase.

View all activity

Organizations

None yet

upvoted a paper 10 days ago

Janus: Disaggregating Attention and Experts for Scalable MoE Inference

Paper • 2512.13525 • Published 12 days ago • 5