This release is a pre-release and may not be stable for production use.
MLCFlow: Simplifying MLPerf Automations
MLCFlow is a versatile CLI and Python interface developed by MLCommons in collaboration with a dedicated team of volunteers (see Contributors). It serves as a streamlined replacement for the CMind tool, designed to drive the automation workflows of MLPerf benchmarks more efficiently. You can use this tool for any of your workflow automation tasks.
The concept behind CMind originated from Grigori Fursin, while the MLPerf Automations project was created by Grigori Fursin and Arjun Suresh, whose collective contributions laid the foundation for modernizing MLPerf benchmarking tools. MLCFlow is a complete replacement for the CMind package with an entirely new implementation led by Arjun Suresh and Anandhu Sooraj, with support from the MLCFlow community. The script automation used in MLCFlow is inherited from the CM project but is now modified to support MLC and provides a simpler interface to the devlopers. Please see the documentation for more details.
On February 9, 2025, MLCFlow released its first stable version, 1.0.0.
Key Features
Building upon the core idea of CMind—wrapping native scripts with Python wrappers and YAML metadata—MLCFlow focuses exclusively on key automation components: Scripts, along with its complementary modules: Cache, Docker, and Experiments. This targeted design simplifies both implementation and interface, enabling a more user-friendly experience.
Status
MLCFlow is now fully equipped for workflow development, with complete support for all previously used CM scripts in MLPerf inference automation. If you're interested in discussions, join the MLCommons Benchmark Infra Discord channel, and check out the latest progress in Issues.
MLC CLI Overview
The MLC Command-Line Interface (CLI) enables users to perform actions on specified targets using a simple syntax:
mlc <action> <target> [options]
Key Components:
<action>: The operation to be performed.<target>: The object on which the action is executed.[options]: Additional parameters passed to the action.
Supported Targets and Actions
1. Repo
- Actions related to repositories, such as cloning or updating.
2. Script
- Manage or execute automation scripts.
3. Cache
- Handle cached data, including cleanup or inspection.
Each target has its own set of specific actions to tailor automation workflows as specified below.
| Target | Action |
|---|---|
| script | run, find/search, rm, mv, cp, add, test, docker-run, show, experiment, doc |
| cache | find/search, rm, show, list, prune, mark-tmp |
| repo | pull, search, rm, list, find/search , add |
MLC started with a compatibility layer where by it supported MLCommons CM automations - Script, Cache and Experiment. Now, MLCFLow has just the Script Automation which is an extension of the Script Automation from CM but with a cleaner integration of Cache Automation and Docker and Test extensions. The old CM scripts are now updated with the latest MLCFlow scripts in the MLPerf Automations repository.
Architectural Diagram
classDiagram
class Action {
+access(options)
+find_target_folder(target)
+load_repos_and_meta()
+load_repos()
+conflicting_repo(repo_meta)
+register_repo(repo_meta)
+unregister_repo(repo_path)
+add(i)
+rm(i)
+save_new_meta(i, item_id, item_name, target_name, item_path, repo)
+update(i)
+is_uid(name)
+cp(run_args)
+copy_item(source_path, destination_path)
+search(i)
}
class RepoAction {
+find(run_args)
+github_url_to_user_repo_format(url)
+pull_repo(repo_url, branch, checkout)
+pull(run_args)
+list(run_args)
+rm(run_args)
}
class ScriptAction {
+search(i)
+rm(i)
+dynamic_import_module(script_path)
+call_script_module_function(function_name, run_args)
+docker(run_args)
+run(run_args)
+test(run_args)
+list(args)
}
class CacheAction {
+search(i)
+find(i)
+rm(i)
+show(run_args)
+list(args)
}
class ExperimentAction {
+show(args)
+list(args)
}
class Index {
+add(meta, folder_type, path, repo)
+get_index(folder_type, uid)
+update(meta, folder_type, path, repo)
+rm(meta, folder_type, path)
+build_index()
}
class Item {
+meta
+path
+repo
+_load_meta()
}
class Repo {
+path
+meta
+_load_meta()
}
class Automation {
+action_object
+automation_type
+meta
+path
+_load_meta()
+search(i)
}
Action <|-- RepoAction
Action <|-- ScriptAction
Action <|-- CacheAction
RepoAction o-- Repo
ScriptAction o-- Automation
CacheAction o-- Index
ExperimentAction o-- Index
CfgAction o-- Index
Index o-- Repo
Index o-- Item
Item o-- Repo
Automation o-- Action
2 maintainers
Release files for mlcflow 1.4.0a5
For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.
Source distribution (sdist)
| File | Size | Uploaded | |
|---|---|---|---|
| mlcflow-1.4.0a5.tar.gz | 244.9 kB | Details |
Built distribution (wheel)
| File | Interpreter | ABI | Platform | Reset |
|---|---|---|---|---|
| mlcflow-1.4.0a5-py3-none-any.whl | Python 3 | none | any | Details |
Total release size: 461.8 kB
Release files / mlcflow-1.4.0a5.tar.gz
| Download URL | mlcflow-1.4.0a5.tar.gz |
|---|---|
| Size | 244.9 kB |
| Tags | Source |
|
SHA-256 checksum How to use checksums |
2aabfbf090fc2bb2e8b6392d558fc5265af27b8062a07f233c9c42257c8cdb81
|
|
BLAKE2b-256 checksum How to use checksums |
86f98f5388f8465c7b21ee61e3721c00d08cdb701792d4e4b07ed48e45bd89f2
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
Yes |
| Uploaded via |
twine/6.1.0 CPython/3.13.13
|
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
PyPI Publish Attestation
PyPI verified that this artifact, at this checksum, originated from the publisher listed below.
Signed by GitHub Actions, verified by PyPI on Sep 11, 2026.
Transparency logRelease files / mlcflow-1.4.0a5-py3-none-any.whl
| Download URL | mlcflow-1.4.0a5-py3-none-any.whl |
|---|---|
| Size | 217.0 kB |
| Tags | Python 3 |
|
SHA-256 checksum How to use checksums |
a82ccbea3e0ec4d60f9624d635859d6b111b84bce4c776c677d60963dabab15b
|
|
BLAKE2b-256 checksum How to use checksums |
4593cd4298c782f6487fd97ae3752de78b54b8c4845d0e2f131f97856c471d4c
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
Yes |
| Uploaded via |
twine/6.1.0 CPython/3.13.13
|
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
PyPI Publish Attestation
PyPI verified that this artifact, at this checksum, originated from the publisher listed below.
Signed by GitHub Actions, verified by PyPI on Sep 11, 2026.
Transparency log
