prime-rl
Project record
Maintained by Prime Intellect12
prime-rl is Prime Intellect's open-source framework for large-scale, asynchronous reinforcement learning and supervised fine-tuning of language models. It separates an FSDP2-based trainer, a vLLM inference service, and an orchestrator that gathers rollouts from verifiers environments, and it supports multi-node deployment on Slurm and Kubernetes.15
- Repository: GitHub repository (external site: github.com)
- License: License (Apache 2.0) (external site: github.com)
- Paper: INTELLECT-3 technical report (arXiv 2512.16144) (external site: arxiv.org)
Availability and license
Overall availability
Documented as available to the general public. Access conditions and license terms may still apply.12
Availability is separate from permission: read the license before using or redistributing.
Public materials checklist
| Item | Status | Notes and evidence |
|---|---|---|
| Source codeIs the source code publicly readable? | Public | Public GitHub repository under the Apache License 2.0.12 |
| DocumentationIs user documentation published? | Public | The repository's docs directory covers architecture, configuration, training, evaluation, scaling, algorithms, and development, and Prime Intellect's documentation site has a prime-rl section.14 |
| Training codeDoes it include code for training models? | Public | Includes trainers for SFT and RL plus evaluation tooling, with optimized training code for selected model families.1 |
| Data informationAre the data it expects or ships with documented? | Public | The README describes training on verifiers environments, installed from the Environments Hub or as opt-in workspace packages, and walks through example tasks such as text reversal, Wordle, and Wikipedia search.1 |
| ReproducibilityAre instructions for reproducing reported results published? | Partial | The README gives setup checks and end-to-end example runs, from small single-GPU tasks to Slurm-based training of larger models. A step-by-step reproduction of the INTELLECT-3 run was not found in the repository.1 |
What it is useful for
Run and use notes
- The README states that at least one NVIDIA GPU is required, that the project is developed and tested on NVIDIA RTX 3090/4090/5090, A100, H100, H200, and B200 GPUs, and that setup uses uv with Python 3.12.1
Organization context
U.S. eligibility
Eligible · basis: U.S.-governed project
The repository is maintained in Prime Intellect's GitHub organization, its README cites Prime Intellect as author, and the INTELLECT-3 report introduces it as Prime Intellect's framework. Prime Intellect, Inc. is a Delaware corporation with U.S. locations (see the prime-intellect record).157
Sources
This listing is not an endorsement, a safety assessment, or a federal approval.