mlperf

Collective Knowledge (CK), Collective Mind (CM/CMX) and MLPerf automations: community-driven projects to learn how to run AI, ML, and other emerging workloads more efficiently and cost-effectively across diverse models, datasets, software, and hardware using MLPerf methodology and benchmarks.

MLCube® is a project that reduces friction for machine learning by ensuring that models are easily portable and reproducible.

Pythonic binding to the Apple Neural Engine

Legacy CM repository with a collection of portable, reusable and cross-platform CM automations for MLOps and MLPerf to simplify the process of building, benchmarking and optimizing AI systems across diverse models, data sets, software and hardware

MLPerf (tm) Tiny Deep Learning Benchmarks for STM32 devices

This repository contains automation scripts designed to run MLPerf Inference benchmarks. Originally developed for the Collective Mind (CM) automation framework, these scripts have been adapted to leverage the MLC automation framework, maintained by the MLCommons Benchmark Infrastructure Working Group.

Benchmark OpenAI-compatible AI endpoints and AI Accelerators in a reproducible structured way

MLCFlow: Simplifying MLPerf Automations

CIFAR10 training repo for MLPerf Tiny Benchmark v0.7

CM interface and automation recipes to analyze MLPerf Inference, Tiny and Training results. The goal is to make it easier for the community to visualize, compare and reproduce MLPerf results and add derived metrics such as Performance/Watt or Performance/$

nvProbe - Open-source NVIDIA GPU benchmark suite for CUDA workload automation, Slurm HPC cluster profiling, and MLPerf reporting

Converting models used by MLPerf Mobile working group to Core ML format

A benchmark suite to used to compare the performance of various models that are optimized by Adlik.

Popperized MLPerf benchmark workflows

Development version of CodeReefied portable CK workflows for image classification and object detection. Stable "live" versions are available at CodeReef portal:

Edge AI deployment knowledge graph — boards, accelerators, ONNX operator kernels and quantized models in one graph. Real ONNX + ONNX Runtime + MLPerf Tiny data plus a generated fleet. Built on Samyama Graph.

Tekton Pipelines to run MLPerf benchmarks on OpenShift

Performance benchmark evidence repository — activates at first-silicon bring-up (Sentinel-1 Q4 2026).

How many decisions/s can one NVIDIA GPU serve under a p99 SLO, and what does a million cost? Reproducible benchmark of the Laya 421M decision models on RTX 2000 Ada, RTX PRO 5000/6000 Blackwell and H100 (+MIG), with TensorRT, torch.compile, ONNX Runtime, and LLM/API baselines.

These are automated test submissions for validating the MLPerf inference workflows