Notes & reading

Blog

Deep dives into systems for large models, plus earlier technical notes and project records.

Source-code reading

01 Megatron-LM 源码精读Parallel training, schedules, model internals 02 lm-evaluation-harness 源码精读Evaluation pipeline, tasks, and model adapters 03 vLLM 源码精读Engine, scheduler, memory, and execution

Articles & projects

Articles1 document
GAN & VAE
Exhibition Robot1 document
Exhibition Robot Q&A
Acceleration & OptimizationCPU · GPU
CPU Optimization GPU Optimization