NVlabs/qerl
QeRL enables reinforcement learning training for 32B parameter LLMs on a single H100 GPU by combining NVFP4 quantization with LoRA.
Not currently ranked — collecting fresh signals.
star history
QeRL is a quantization-enhanced reinforcement learning framework for large language models that addresses the resource intensity of RL training. It combines NVFP4 quantization with Low-Rank Adaptation (LoRA) to accelerate the rollout phase of RL and reduce memory overhead, enabling training of 32B parameter models on consumer-grade hardware.
Frequently asked
- What is NVlabs/qerl?
- QeRL enables reinforcement learning training for 32B parameter LLMs on a single H100 GPU by combining NVFP4 quantization with LoRA.
- Is qerl open source?
- Yes — NVlabs/qerl is open source, released under the Apache-2.0 license.
- What language is qerl written in?
- NVlabs/qerl is primarily written in Python.
- How popular is qerl?
- NVlabs/qerl has 527 stars on GitHub.
- Where can I find qerl?
- NVlabs/qerl is on GitHub at https://github.com/NVlabs/qerl.