~/mlsys

Notes from below the framework line — CUDA, compilers, LLVM/MLIR, quantization, and LLM inference systems.

2026

03-30 What Happens to FlashAttention When Vera Rubin Ships? #hardware
03-30 FlashAttention-4: When Tensor Cores Got Too Fast for Everything Else #hardware
02-23 Spark, Cerebras, and the Future of Low-Latency AI Inference #hardware
02-15 MLIR Is Not Just Another IR #compiler
01-19 vLLM and PagedAttention: Why KV Cache Management Matters #inference
01-16 LLM Serving 101: Prefill, Decode, Batching, and the Systems Behind Large Language Models #inference

2025

09-20 MXFP4 in GPT-OSS : Why Everyone Talks About It #quantization
06-06 Context-Free Grammar (CFG) #compiler
06-04 Backus Naur Form (BNF) #compiler
06-02 Compiler Basic Structure #compiler
05-31 Postfix Notation #compiler
05-07 LLVM-Flow (OSS) #llvm
05-05 LLVM-Block (OSS) #llvm
04-29 History of LLVM #llvm
04-27 Debug & Metadata #llvm
04-26 Optimization Pass #llvm
04-25 Static Single Assignment (SSA) #llvm
04-24 Basic Block & CFG #llvm
04-22 LLVM IR Syntax #llvm
04-21 LLVM IR #llvm
04-20 LLVM Basic Structure #llvm
02-13 Host-Device Synchronization #cuda
02-09 Kernel Configuration #cuda
02-05 Advanced Atomic Operations #cuda
01-30 Basic Atomic Operations #cuda
01-25 Bank Conflict #cuda
01-20 Memory Alignment and Coalescing #cuda
01-17 Thread Hierarchy #cuda
01-15 CPU vs. GPU Architecture #cuda
01-12 Intro to CUDA #cuda