Description
grpo-rlvr-training is an agent skill from the wshobson/agents collection for training reasoning and verifiable-task behavior in language models with GRPO (Group Relative Policy Optimization) and reinforcement learning from verifiable rewards (RLVR). It applies when task success can be checked algorithmically, such as a math answer that is right or wrong, or code that passes its tests, so the model can learn from a reward signal instead of from demonstrations or preference pairs. The skill is part of the wshobson fine-tuning sequence. It assumes the routing step has already decided that reinforcement learning is the right method, and it points to the separate skills for demonstration-based and preference-based training. The skill covers when RL is the right tool, a reference recipe for running GRPO, a mandatory gate that inspects the reward function before a run, and how to choose a GRPO variant when the base recipe behaves badly. The reward inspection step matters because a model will optimize whatever the reward measures, including shortcuts that do not reflect the real goal. It is installed with the skills CLI from the wshobson/agents repository. Its skills.sh page showed about 2.4K installs. The security audits showed Gen Agent Trust Hub with a warning, Socket passing and Snyk passing at the time of the scan.…
Highlights
- GRPO and RLVR training for verifiable tasks
- Reference training recipe
- Mandatory reward-inspection gate
- Choosing a GRPO variant
- Part of the fine-tuning sequence
- Installs with the skills CLI
Ratings & Reviews
No ratings yet
Be the first to rate grpo-rlvr-training.
People also visit
grpo-rlvr-training FAQ
When is GRPO the right method?
When task success has a verifiable pass or fail signal, such as an answer that can be checked automatically.
What is the reward-inspection gate?
A required check of the reward function before training, because a model optimizes what the reward measures, including shortcuts.
Which methods are the alternatives?
Demonstration-based training covered by lora-qlora-recipes, and preference-based training covered by preference-optimization.
Should I start with this skill?
The skill assumes the method was already routed here by finetuning-method-selection, so start there if you are unsure.
How do I install it?
Run npx skills add https://github.com/wshobson/agents --skill grpo-rlvr-training.
What security audits does it have?
skills.sh lists three audits: Gen Agent Trust Hub Warn, Socket Pass and Snyk Pass. A Warn or Fail is worth reading before you install.
What are the best grpo-rlvr-training alternatives?
Popular alternatives to grpo-rlvr-training include lora-qlora-recipes, preference-optimization and dataset-curation. See all grpo-rlvr-training alternatives