pytorch/kineto

A CPU+GPU Profiling library that provides access to timeline traces and hardware performance counters.

What it solves

Kineto provides performance observability and diagnostics for machine learning workloads, helping developers identify and resolve bottlenecks in common ML components.

How it works

It functions as a profiling library, with its core component, Libkineto, focusing on low-overhead GPU timeline tracing. It is integrated directly into the PyTorch Profiler to provide system-level profiling and actionable recommendations for performance issues.

Who it’s for

ML engineers and developers using PyTorch who need to analyze GPU performance and diagnose system-level bottlenecks in their models.

Highlights

  • Low-overhead GPU timeline tracing via Libkineto.
  • Integration with external system-level profiling tools.
  • Compatibility with popular visualization platforms and analysis pipelines.
  • Provides actionable recommendations for common performance issues.

Related

  • Project
    meta-pytorch/tritonbench

    A collection of PyTorch operators used to evaluate the performance of Triton and its integration with PyTorch.

  • Project
    pytorch/ao

    A PyTorch-native library for model optimization that provides quantization and sparsity tools to accelerate training and inference while reducing memory usage.

  • Project
    meta-pytorch/tnt

    A library of training tools and utilities for PyTorch users to streamline the model training process.

  • Project
    facebookresearch/pytorch3d

    A PyTorch-based library for 3D Computer Vision research providing differentiable rendering and efficient tools for manipulating 3D meshes and point clouds.

  • Project
    facebookincubator/dynolog

    Dynolog is a lightweight performance monitoring daemon for heterogeneous CPU-GPU systems, designed to help debug bottlenecks in large-scale AI training. It provides always-on monitoring of CPU, GPU, and kernel metrics, plus on-demand remote profiling of PyTorch workloads across many GPUs.