# combust/mleap

MLeap: Deploy ML Pipelines to Production

Repository: https://github.com/combust/mleap
Canonical: https://ross.abutalabs.com/products/mleap
Homepage: https://combust.github.io/mleap-docs/
Language: Scala
License: Apache-2.0
License Family: permissive
Topics: scikit-learn, spark, data-pipelines, transformers, tensorflow, scala, python
Last push: 2026-07-21T17:38:04+00:00

## Health v2 (maintenance only)
Score: 95/100 (v2, computed 2026-09-02T17:46:02.011165+00:00)
- activity 93, release rhythm 94, longevity 100
- inputs: {"age_days": 3662, "days_push": 43, "days_rel": 43, "gap_med": 5.0, "n_releases_24m": 9}
- flags: none
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1543, forks 316 (observed 2026-08-28T04:05:01.278990+00:00)

## What it is
MLeap is a serialization format (Bundle.ML) and portable execution engine for machine learning pipelines, implemented in Scala with Python support. It lets you train pipelines in Spark, Scikit-learn, or TensorFlow, export them once, and run them on a lightweight JVM runtime without dependencies on Spark, sklearn, numpy, or pandas.

## Use cases
- deploy spark ml pipelines to production without a spark cluster
- serve scikit-learn models from a jvm api service
- export trained ml pipelines to a portable format for realtime scoring
- run ml feature transformers and models without python dependencies
- serialize ml pipelines as  or protobuf bundles
- mix spark and scikit-learn trained components in one deployable pipeline
- score models in batch mode by deserializing bundles back into spark

## When to choose
- you train models in Spark or Scikit-learn but serve them from JVM-based services
- you need low-latency realtime scoring without Spark runtime overhead
- you want portable model bundles not locked to a cloud ML platform
- your production stack is JVM-based and cannot depend on Python

## When to avoid
- your models are deep learning focused with heavy TensorFlow/PyTorch serving needs better served by dedicated serving stacks
- your entire stack is Python and you can serve sklearn models directly
- you rely on custom transformers not supported by MLeap's transformer set
- you need GPU-accelerated inference

## Facets
- artifact type: library
- maturity: active
- function: serialization, machine-learning, etl
- domain: machine-learning, big-data, developer-tools
- platform: jvm, python, cross-platform
- tags: spark, scikit-learn, tensorflow, ml-pipelines, model-deployment, bundle-ml, protobuf, realtime-scoring, data-engineering

## Member repositories
- combust/mleap (main) score 95

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:05:01.278990+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T04:30:40.933342+00:00, confidence not recorded.
  - readme: https://github.com/combust/mleap (fetched 2026-08-28T04:05:01.278990+00:00, sha 9a910b55a4a3)
  - homepage: https://combust.github.io/mleap-docs/ (fetched 2026-08-29T11:31:54.354757+00:00, sha 7e990b4194d0)
- Data as of 2026-08-30T08:39:29.467469+00:00.