yifan cui

projects & interests

• vLLM
  High-throughput and memory-efficient inference engine for LLMs
  python · cuda · pytorch
• TileLang
  DSL for high-performance GPU/CPU/Accelerator kernels
  python · compiler
• SGLang
  High-performance serving framework for LLMs and multimodal models
  python · inference
• DeepGEMM
  Clean and efficient FP8 GEMM kernels with fine-grained scaling
  cuda · fp8
• FlashMLA
  Efficient MLA decoding kernels
  c++ · cuda
• nanochat
  Train a 500M+ model end-to-end for less than $100
  LLM training