mLeap
mLeap is an open-source serialization framework and execution engine for machine learning pipelines, originally developed by Combust, a data engineering consultancy.
Profile
mLeap serializes machine learning pipelines from Spark, Scikit-learn, and TensorFlow into a portable format and executes them in production without the original training framework.
mLeap is an open-source serialization framework and execution engine for machine learning pipelines, originally developed by Combust, a data engineering consultancy. The project, hosted on GitHub under the combust/mleap repository, has accumulated over 1,500 stars and 317 forks as of June 2026, with 1,080 commits and contributions from Yelp engineers. mLeap enables users to export trained ML pipelines from Apache Spark, Scikit-learn, and TensorFlow into a portable, serialized format (MLeap Bundle) and execute them in production without dependencies on the original training frameworks. The project's last significant commit, in March 2026, upgraded support for PySpark 4.0.1, Scala 2.13.16, Java 17, and XGBoost 2.0.3, reflecting ongoing maintenance. mLeap does not appear to have a corporate entity, dedicated funding, or disclosed revenue; it remains a community-maintained open-source tool.
Its primary users are data engineering teams at companies like Yelp, which need to deploy Spark-trained models into low-latency, JVM-based production environments. The project's market position is niche but durable, competing with tools like ONNX and PMML for model portability. However, mLeap's reliance on volunteer contributions and its narrow focus on JVM ecosystems limits its growth.
There is no evidence of recent funding rounds, acquisitions, or layoffs related to mLeap; the project's development pace has slowed, with only a handful of commits in 2025-2026. The most notable recent activity is a March 2026 pull request from a Yelp engineer fixing GPG signing for releases, indicating corporate usage but not commercial backing.
Who buys this
- Data engineering teams deploying Spark ML pipelines into production
- Organizations using JVM-based infrastructure for model serving
- Teams needing to decouple model training from inference environments
- Companies with legacy Spark ML workflows seeking lightweight deployment
Strengths and what to watch
Strengths
- Lightweight execution engine that runs ML pipelines without Spark or Python dependencies, reducing infrastructure overhead.
- Active maintenance with support for modern versions of Spark (4.0.1), Scala (2.13.16), and Java 17, as shown in recent commits.
- Proven adoption by Yelp, with a Yelp engineer contributing a fix for release signing in March 2026, indicating real-world use.
Watch for
- No corporate backing or dedicated funding; the project relies on volunteer contributions and may face stagnation if key maintainers leave.
- Limited ecosystem compared to ONNX or PMML; mLeap's focus on JVM and Spark may alienate Python-first teams.
- Slow development pace: only a few commits in 2025-2026, with the last significant upgrade in November 2025, suggesting reduced community activity.
Recent moves
Key Information
- Industry
- MLOps & AI Infra
- Founded
- 1986
Frequently Asked Questions
What is mLeap and what does it do?
mLeap is an open-source serialization framework and execution engine for machine learning pipelines. It exports trained models from Spark, Scikit-learn, and TensorFlow into a portable format and runs them in production without needing the original training frameworks.
How does mLeap help deploy Spark ML models?
mLeap serializes Spark ML pipelines into a lightweight bundle that executes on a JVM without Spark dependencies. This reduces infrastructure overhead and allows low-latency inference in production environments, as used by companies like Yelp.
Which ML frameworks does mLeap support?
mLeap supports Apache Spark, Scikit-learn, and TensorFlow. It can export trained pipelines from these frameworks into a portable MLeap Bundle format and execute them independently, with recent updates adding support for PySpark 4.0.1 and XGBoost 2.0.3.
Is mLeap actively maintained and who uses it?
mLeap is community-maintained with no corporate backing. It has over 1,500 GitHub stars and recent commits from Yelp engineers, including a March 2026 fix for GPG signing. Development has slowed, with only a few commits in 2025-2026.
How does mLeap compare to ONNX and PMML?
mLeap competes with ONNX and PMML for model portability but focuses narrowly on JVM and Spark ecosystems. It is lighter for JVM deployments but has a smaller ecosystem and slower development, which may limit adoption compared to broader alternatives.
What are the main limitations of mLeap?
mLeap relies on volunteer contributions with no dedicated funding, risking stagnation if maintainers leave. Its narrow focus on JVM and Spark may not suit Python-first teams, and its development pace has slowed, with only a handful of commits in recent years.
Sources
- github.com — Project repository details: 1.5k stars, 317 forks, 1,080 commits, last commit March 2026, Yelp contributor, support for Spark 4.0.1, Scala 2.13.16, Java 17, XGBoost 2.0.3.
- news.crunchbase.com — Confirms no connection between mLeap and Magic Leap; Magic Leap is an AR startup with $2.3B in funding, unrelated to mLeap.
- onegiantleap.com — Confirms no connection between mLeap and LEAP conference; LEAP is a Saudi tech conference, unrelated to mLeap.
- investors.leaptx.com — Confirms no connection between mLeap and Leap Therapeutics; Leap Therapeutics is a biotech subsidiary of Cypherpunk Technologies, unrelated to mLeap.
- www.fiercebiotech.com — Confirms no connection between mLeap and Leap Therapeutics; Leap Therapeutics laid off 75% of staff in June 2025, unrelated to mLeap.