Inside AI SystemsJul 238 min read
Multi-Head Attention To Flash Attention
What actually changes when you go from manual attention to a fused kernel
Efficient Inference
Quantization, pruning, distillation, speculative decoding, and serving large models cheaply.
2 articles
Inside AI SystemsJul 213 min read
Prefill And Decode
Understanding Prefill and Decode in LLM Inference
