Privacy and advertising choices
Git-Stars uses essential storage for site operation. Optional analytics and ad-measurement scripts stay disabled unless you accept them; partners such as Google may then use cookies or similar identifiers where required. Privacy Policy
Agentic RL Training at Scale
This repository page has enough original analysis, source evidence, and comparison context to be treated as an indexable Git-Stars review.
100
review score
Problem solved
It addresses the challenge of training large-scale agentic RL models efficiently by providing a fully asynchronous, high-throughput framework that scales to 1000+ GPUs. It solves the complexity of managing distributed training with FSDP2, vLLM inference, and multi-node deployment, while supporting a wide range of model families including MoE and multimodal models.
Deployment reality
The available setup signal starts with: curl -sSL https://raw.githubusercontent.com/PrimeIntellect-ai/prime-rl/main/scripts/install.sh | bash. Treat this as a starting point, then ask Codex or Claude to inspect the README, environment variables, runtime version, and deployment target before production use.
Commercial use
The recorded license is Apache-2.0. This is a useful commercial-use signal, but teams should still verify license text, dependencies, model/API terms, and trademark constraints.
Capability ceiling
Developers can build and train large-scale RL agents for tasks like software engineering (SWE), agentic environments, and multimodal reasoning. The framework supports end-to-end post-training (SFT, RL, evals), and can handle frontier models like GLM-5 with advanced features like FP8 inference, PD disaggregation, and Mooncake KV offload. The ceiling includes training 1T+ parameter MoE models on 1000+ GPUs with expert and context parallelism.
Git-Stars is independent and not affiliated with GitHub or this project. Analysis may be AI-assisted and based on public repository metadata plus short README-derived summaries. We do not mirror full README files, docs, issues, or social comments.
PrimeIntellect-ai/prime-rl is tracked as a Python project in the AI Agent area. This evaluation combines public GitHub metadata, category signals, short source summaries, and Git-Stars editorial rules rather than copying project documentation.
Momentum check: the repository has 1.9k total stars, with +0 today, +79 this week, and +0 this month. These growth windows help distinguish durable adoption from short-lived visibility spikes.
Maintenance check: current activity is Active; the latest push was 9 days ago, and the open issue queue is 183, about 9.54% of total stars. Treat this as an adoption signal, not a substitute for engineering due diligence.
Adoption check: 401 forks and 1.9k watchers suggest how often the project is reused or followed. License signal: Apache-2.0. Always verify license compatibility before commercial or internal use.
Practical fit: this project is most relevant when you need Python teams evaluating ecosystem-native tooling. Be cautious when you have low tolerance for large unresolved issue queues.
Source check: Git-Stars currently has 2 explicit source reference(s) for this report and a recent growth signal of 79. Follow the original GitHub repository for final install, security, and release information.
Analysis
ReadyOriginal problem, capability, and verdict guidance are available.
Sources
ReadyRepository metadata and README/source references are attached.
License
ReadyApache-2.0 is recorded for review.
Maintenance
ReadyRecent activity is visible in repository metadata.
Alternatives
ReadyEnough nearby projects exist for comparison.
Momentum
1.9k Stars
Reuse
401 Forks
Attention
1.9k Watchers
Maintenance
active
License
Apache-2.0
Open issues
183
prime-rl is a framework for large-scale reinforcement learning, designed to be easy to use and hackable while scaling to 1000+ GPUs. It supports fully asynchronous RL training, end-to-end post-training (SFT, RL, evals), and integrates with verifiers environments.
Key Features
- Fully asynchronous RL for high-throughput agentic training at scale. - Performant: built to train 1T+ MoE models on 1000+ GPUs with FSDP2 and vLLM, including FP8 inference, PD disaggregation, EP and CP parallelism. - Native integration with verifiers environments, end-to-end post-training (SFT, RL, evals), multi-node deployment with Slurm/Kubernetes, and multimodal support for VLMs.
AI Agent
Agent frameworks, autonomous workflows, and tool-use systems
curl -sSL https://raw.githubusercontent.com/PrimeIntellect-ai/prime-rl/main/scripts/install.sh | bash73
Health Score
Active
Commit Activity
Feb 18, 2025
Created
Aug 15, 2026
Last push
+0
Today's growth
+0
7-day growth
+0
30-day growth
Forks
Open
Watchers
PrimeIntellect-ai
GitHub profileobra/superpowers
An agentic skills framework & software development methodology that works.
affaan-m/ECC
The agent harness performance optimization system. Skills, instincts, memory, security, and research-first development for Claude Code, Codex, Opencode, Cursor and beyond.
NousResearch/hermes-agent
The agent that grows with you
mattpocock/skills
Skills for Real Engineers. Straight from my .agents directory.
anomalyco/opencode
The open source coding agent.
✓
License
✓
Forked
✓ Active
Maintained
Problem Solved
It addresses the challenge of training large-scale agentic RL models efficiently by providing a fully asynchronous, high-throughput framework that scales to 1000+ GPUs. It solves the complexity of managing distributed training with FSDP2, vLLM inference, and multi-node deployment, while supporting a wide range of model families including MoE and multimodal models.
Capabilities
Developers can build and train large-scale RL agents for tasks like software engineering (SWE), agentic environments, and multimodal reasoning. The framework supports end-to-end post-training (SFT, RL, evals), and can handle frontier models like GLM-5 with advanced features like FP8 inference, PD disaggregation, and Mooncake KV offload. The ceiling includes training 1T+ parameter MoE models on 1000+ GPUs with expert and context parallelism.
Bottom Line
This framework is ideal for researchers and organizations that need to train large-scale RL agents at scale, especially those working with MoE or multimodal models. It is not suitable for those without access to NVIDIA GPUs or who need a simple, low-resource setup. The key trade-off is its complexity and hardware requirements in exchange for extreme scalability and performance.