NVIDIA-NeMo/Gym

Health v2 · maintenance only

Adoption not part of the score

1156 stars · 304 forks observed · 2026-09-03

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-30, confidence not recorded

NeMo Gym is a Python library from NVIDIA for evaluating and improving LLM models and agents using environments. It provides infrastructure to build environments (dataset, agent harness, verifier, state), run evaluation and RL training at scale, and ships a hub of popular benchmarks and training environments.

Use cases

• evaluate llm agents in stateful environments
• run rl training environments for language models
• benchmark models on tool calling and code execution tasks
• build custom environments with verifiers for agent training
• scale agent evaluation to thousands of concurrent rollouts
• integrate rl frameworks with agent environments
• reproducible agent benchmarking across teams

When to choose

• you need stateful evaluation of models or agents with verifiers
• you want to scale evaluation or training to thousands of concurrent environments
• you want to transition seamlessly between evaluation, agent optimization, and RL training
• you want a shared environment hub of popular benchmarks

When to avoid

• you just need a stateless check on model outputs - a simple script suffices
• you need a general-purpose game RL gym rather than LLM agent environments
• you are not working with LLMs or agents

Facets

library · maturity active

reinforcement-learning agent-framework benchmarking llm-training testing mcp reinforcement-learning large-language-models machine-learning developer-tools python cli rl-environments agent-evaluation verifiers llm-benchmarks nvidia-nemo agent-harnesses sandboxing ai-agents linux docker gpu

Member repositories

For agents

markdown · JSON · MCP: product_card(name="NVIDIA-NeMo/Gym")