Chakra is an open and interoperable graph-based representation of AI/ML workloads focused on enabling and accelerating AI SW/HW co-design. Chakra execution traces represent key operations, such as compute, memory, and communication, data and control dependencies, timing, and resource constraints.
This is a repository of Chakra schema and a complementary set of tools and capabilities to enable the collection, analysis, generation, and adoption of Chakra execution traces by a broad range of simulators, emulators, and replay tools.
Chakra is under active development as a MLCommons® research project. Please see MLCommons Chakra Working Group for more details for participating in this effort.
Check out USER_GUIDE for details.
Please fine following useful resources about Chakra:
A detailed description of the original motivation and guiding principles can be found here. The paper was published prior to Chakra becoming a MLCommons project. Please cite the following paper when referring to the latest Chakra schema and tools.
@inproceedings{sridharan2026mlcommonschakra,
title = {MLCommons Chakra: Advancing Performance Benchmarking and Co-design using Standardized Execution Traces},
author = {Sridharan, Srinivas and Balogh, Andy and Beckmann, Bradford M. and Coutinho, Brian and Feng, Louis and Fu, Sheng and Gao, Sanshan and Garakani, Mehryar and Heo, Taekyung and Kanter, David and Ladd, Josh and Li, Ziwei and Liu, Winston and Man, Changhai and Mihailescu, Dan and More, Spandan and Park, Joongun and Ramachandran, Ashwin and Ramakrishnaiah, Vinay and Rashidi, Saeed and Reddi, Vijay Janapa and Sharma, Puneet and Tian, Phio and Won, William and Wu, Hanjiang and Xu, Huan and Yoo, Jinsun and Krishna, Tushar},
booktitle = {Proceedings of the Ninth Annual Conference on Machine Learning and Systems},
year = {2026}
}Chakra is released under the MIT license. Please see the LICENSE.md file for more information.
We actively welcome your pull requests! Please see CONTRIBUTING.md for more info.