Marin
Project record
Maintained by Open Athena (Open Athena AI Foundation Inc.)1910, Stanford Center for Research on Foundation Models (CRFM)27
Marin is an open lab and software platform for developing foundation models, covering data curation, filtering, tokenization, pretraining, post-training, and evaluation, with experiments, code, and training logs published as the work happens. It began at Stanford in 2024, originating in Stanford CRFM and HAI, and is now primarily developed by the nonprofit Open Athena with Stanford CRFM as a core collaborator. Models trained with it include Marin 8B and Marin 32B.2158
- Website: Website (external site: marin.community)
- Repository: Repository (external site: github.com)
- Documentation: Documentation (external site: marin.readthedocs.io)
- Model hub: Models and datasets (Hugging Face) (external site: huggingface.co)
- License: LICENSE (Apache 2.0) (external site: github.com)
- Documentation: Marin 8B retrospective (external site: github.com)
Availability and license
Overall availability
Source code and documentation are public. Model checkpoints are published separately on Hugging Face; the Marin 8B Base repository is not gated.24713
Availability is separate from permission: read the license before using or redistributing.
The Apache 2.0 entry covers the Marin repository's code. Model releases carry their own terms: the Marin 8B Base model card states that its code and model are released under Apache 2.0, and that three of the Markdown-converted datasets Marin made for training (StackExchange, Wikipedia, and one Ar5iv set) keep the licenses of their original sources. This project record does not assess individual model releases.37
Component reuse rights
- weights
- Unknown — no complete fact-level rights review
- code
- Reviewed qualifying license recorded — check scope and conditions
- data
- Unknown — no complete fact-level rights review
- documentation
- Unknown — no complete fact-level rights review
No complete system-rights review is recorded for this release.
Public materials checklist
| Item | Status | Notes and evidence |
|---|---|---|
| Source codeIs the source code publicly readable? | Public | Public GitHub repository under Apache 2.0.23 |
| DocumentationIs user documentation published? | Public | Documentation on Read the Docs and in the repository's docs folder includes tutorials, explanations, experiment reports, a developer guide, and technical reference.42 |
| Training codeDoes it include code for training models? | Public | The repository contains the pretraining and post-training pipeline, and the README links the training script used for Marin 8B. The Marin 8B model card points to the Levanter training library.27 |
| Data informationAre the data it expects or ships with documented? | Public | The Marin 8B retrospective documents the data mixture in each training phase (for example DCLM Baseline, StarCoder, Proofpile 2, Nemotron-CC, and Marin-made Markdown corpora), and the README links pipelines that reproduce the Delphi training mixture from public datasets.62 |
| ReproducibilityAre instructions for reproducing reported results published? | Public | Experiments are declared in code as dependent steps, and the documentation walks through installation and example training runs. The 8B retrospective records hardware, phases, and hyperparameters and links its experiment script and Weights & Biases report. Reproducing the large runs requires TPU or GPU clusters.246 |
What it is useful for
Run and use notes
- The Marin 8B retrospective reports training on two TPU v5e-256 slices for the first phase and on a TPU v4-2048 slice for later phases, with parameters and optimizer states in float32 and computation in bfloat16.6
Organization context
U.S. eligibility
Eligible · basis: U.S.-governed project
The Marin site says the project is primarily developed by Open Athena, and the README names Stanford CRFM and Open Athena as its core collaborators. Open Athena's terms of use name Open Athena AI Foundation Inc., at 1245 Broadway in New York, under New York law; the IRS extract for New York lists Open Athena Foundation Inc. at the same address as a 501(c)(3) corporation. The Marin 8B model card says the model was developed by the Marin team at Stanford CRFM. Both documented stewards are U.S.-based; the slight difference between the two Open Athena entity names is resolved here only by the matching address.121011127
Sources
This listing is not an endorsement, a safety assessment, or a federal approval.