paper-with-me

홈 › Papers

Accelerating a Triton Fused Kernel for W4A16 Quantized Inference with SplitK work decomposition

2024-01-05 · Adnan Hoque, Less Wright, Chih-Chieh Yang, Mudhakar Srivatsa, Raghu Ganti

We propose an implementation of an efficient fused matrix multiplication kernel for W4A16 quantized inference, where we perform dequantization and GEMM in a fused kernel using a SplitK work decomposition. Our implementation shows improvement for the type of skinny matrix-matrix multiplications found in foundation model inference workloads. In particular, this paper surveys the type of matrix multiplication between a skinny activation matrix and a square weight matrix. Our results show an average of 65% speed improvement on A100, and an average of 124% speed improvement on H100 (with a peak of 295%) for a range of matrix dimensions including those found in a llama-style model, where m < n = k.

📄 PDF Abstract BibTeX arXiv:2402.00025

Code (0)

등록된 구현이 없습니다.

Methods 이 논문이 사용한 방법론

SPEED The monocular depth estimation (MDE) is the task of estimating depth from a single frame. This information is an essential knowledge in many computer vision tasks such as scene…

Similar Papers 제목 키워드 기반

Geak: Introducing Triton Kernel AI Agent & Evaluation Benchmarks

2025-07-31 · Jianghui Wang, Vinay Joshi, Saptarshi Majumder, Xu Chao 외 arxiv

The demand for AI-generated GPU kernels is rapidly growing, influenced by the need for scalable, hardware-optimized solutions in both industry and academia. As deep learning workloads grow in complexity and diversity, it…

Code Generation

rl-triton: High-Performance Triton GPU Kernels for Reinforcement Learning Credit Assignment

2026-08-18 · Lars Simon Zehnder arxiv

We present rl-triton, an open-source library of high-performance GPU kernels for reinforcement learning credit assignment, implemented in Triton. The core contribution is a unified associative scan framework that recasts…

Reinforcement Learning

Evaluating CUDA Tile for AI Workloads on Hopper and Blackwell GPUs

2026-04-25 · Divakar Kumar Yadav, Tian Zhao, Deepak Kumar arxiv

NVIDIA's CUDA Tile (CuTile) introduces a Python-based, tile-centric abstraction for GPU kernel development that aims to simplify programming while retaining Tensor Core and Tensor Memory Accelerator (TMA) efficiency on m…

Compiler-First State Space Duality and Portable $O(1)$ Autoregressive Caching for Inference

2026-03-10 · Cosmo Santoni, Anmol Thapar arxiv

High-throughput Mamba-2 inference is usually tied to fused CUDA and Triton kernels, limiting portability across accelerator backends. We show that the state space duality (SSD) recurrence has a compiler-friendly structur…

Diagonal-Tiled Mixed-Precision Attention for Efficient Low-Bit MXFP Inference

2026-04-05 · Yifu Ding, Xinhao Zhang, Jinyang Guo arxiv

Transformer-based large language models (LLMs) have demonstrated remarkable performance across a wide range of real-world tasks, but their inference cost remains prohibitively high due to the quadratic complexity of atte…