Domains
Forms
Mechanism
Breaks down how the skill works and what it produces, so you can quickly judge whether it fits your scenario and value.
When task success is algorithmically checkable (math, code, tool calls, structured output), when designing GRPO reward functions, or when a GRPO run diverges or reward-hacks.
A validated GRPO config (kwarg values + reward functions) consumed directly by a training engineer, reducing reward-hacking risk and producing reasoning models with verifiable behavior.
Use Cases
A team builds a math tutor LLM where answers are gradeable by exact match. They use this skill to configure GRPO with a correctness+format reward, inspect 80 sampled outputs against the reward function pre-training, and catch a bug where malformed answers scored equal to correct ones. Fixing it yields a stable run that improves accuracy on held-out problems without reward-hacking.
An engineer sees GRPO training collapse entropy on long chain-of-thought. Following the skill's variant table, they switch to DAPO to relax KL penalty and decouple clip bounds. The run recovers exploration, produces longer valid reasoning, and avoids degenerate outputs — all without preemptively picking a variant before the failure appeared.
Skill Relationships
Dependency relationships read as "the upper tier points to the lower tier." The current Skill sits in the middle tier — above are Skills that depend on it, below are Skills it depends on.
Tier 1 · These Skills Use Me
Tier 2 · Current Skill
Tier 3 · I Use These Skills
Browse this skill's relationships within its skillset. Click a node to switch the side panel; use the search box to jump to any skill.
Skill File
