kernel-fusion

Burn is a next generation tensor library and Deep Learning Framework that doesn't compromise on flexibility, efficiency and portability.

AMD RAD's multi-GPU Triton-based framework for seamless multi-GPU programming

An agent harness that compiles a model into one provably-correct, self-retargeting CUDA megakernel and self-tunes it past cuBLAS at batch-1 LLM decode, paper: https://arxiv.org/abs/2606.09682

An efficient concurrent graph processing system

GoPTX: Fine-grained GPU Kernel Fusion by PTX-level Instruction Flow Weaving

Fused Triton kernels for TurboQuant KV cache compression — 2-4 bit quantization with RHT rotation. Drop-in HuggingFace & vLLM integration. Up to 4.9x KV cache compression for Llama, Qwen, Mistral, and more.

A complete GPT built entirely on NVIDIA cuTile: declarative GPU kernels in ~20 lines instead of ~150 hand-tuned CUDA, matching PyTorch speed in a ~10MB footprint

Noeris — autonomous kernel fusion discovery + Triton autotuning for LLM kernels and Gemma layer deeper fusion (A100/H100 wins).

LAMB go brrr

面向 Transformer 推理的可验证 Triton 融合算子与 torch.library 集成

MLX + Metal implementation of mHC: Manifold-Constrained Hyper-Connections by DeepSeek-AI.

A CUDA framework for separating graph-recoverable launch overhead from byte-elimination gains in LLM decode-chain fusion

Browser-native quantum chemistry in WebGPU and TypeScript. HF, UHF, DFT, MP2, CCSD, unrestricted CCSD and CCSD(T), with WGSL kernels for the CCSD(T) triples and the Fock build. EE/IP/EA-EOM-CCSD, Cholesky density fitting, statevector and MPS/DMRG. Cross-checked against PySCF, ITensor and brute-force Hamiltonians.

Assigment 3 for the "Parallel & Distributed Systems" course (ECE, AUTh) - Fall 2024

Write the math. Get the kernel. Fused CUDA kernel generation from mathematical specifications.

Fused normalisation CUDA kernels for PyTorch — residual-add+LayerNorm/RMSNorm, fp8 outputs, fast deterministic backward, torch.compile-native, with reproducible A100/H100 benchmarks

GPU fusion code and algorithm

Compile time kernels fusion and expression trees as Alpaka boost.odeint backend. This is my team project developed in collaboration with and under the supervision of HZDR.

Ahead-of-time compiler for static-shape PyTorch inference. Fuses elementwise and reduction chains into Triton kernels, plans memory into one arena, captures CUDA graphs. 3x faster than torch.compile at batch 1.

Fused STFT + mel filterbank CUDA kernel — ~3x faster log-mel spectrograms than torchaudio's GPU path, verified on NVIDIA GB10 (DGX Spark)