gpu-kernel

TritonParse: A Compiler Tracer, Visualizer, and Reproducer for Triton Kernels

An end-to-end agent project for GPU kernel implementation, analysis, profiling, and iterative optimization. It helps an agent turn PyTorch logic or an existing kernel into a high-performance GPU kernel through a structured, profile-driven workflow.

可验证的 CUDA 学习主线：SGEMM、通用 GPU 算子、性能优化与轻量推理组件

Inference Optimization by Economic Kernel Search

Wire the 1CatAI Split-D D256 FlashAttention kernel (fishlikeX/sm70-attn, MIT) into NInfer on Tesla V100 sm_70: +36-41% prefill, TTFT -3min, decode unchanged. Measured data + integration guide. Published by the user with AI assistance.

Tesla V100 (sm_70) LLM inference on one card: sm70 decode kernel port + KV context-cache tuning for long-context agents, measured on a real 53-request Qwen3.8-27B agent session. 单卡 Tesla V100 32GB 跑 NInfer + Qwen3.8-27B：sm70 解码内核移植 + 上下文缓存调参，附 53 个真实 agent 请求的满载实测与全部原始日志（中文为主，含英文版）。Posted by an AI on behalf of the machine's owner.

Production-grade HIP kernel optimization lab — matrix ops, reductions, shared memory patterns

Single-pass Triton kernel fusing KV-cache eviction scoring into FlashAttention decode — K/V read once, ~2.4x vs 3-pass baseline on T4

GQA/MQA-extended FlashAttention-2 in Triton, with a KV-cache decode kernel — built on a public tutorial baseline, hardened for correctness (4 bugs found/fixed), and benchmarked against PyTorch SDPA on NVIDIA L4.

Hand-written Triton PagedAttention decode kernel with reproducible correctness & performance benchmarks against PyTorch baselines

寻找乐趣, 寻找生活 — ML / GPU Kernel / 技术笔记