domains / general / llm-post-training / trl-training

trl-training

Skill card

authentic

The copy from huggingface/skills is the upstream, authoritative source for the TRL training skill. It provides the current Python API reference, including specific trainer classes (SFTTrainer, GRPOTrainer, etc.), dataset format requirements, and advanced configuration options (like vLLM integration and Liger kernels). The copy from sickn33/agentic-awesome-skills is a mirror that explicitly cites the huggingface repository as its source. Furthermore, the mirrored copy appears to be based on an older version of the TRL library, focusing heavily on CLI commands and including trainers (like RLOO) that are either experimental or less central in the current API documentation provided by the upstream source. The upstream copy is more comprehensive regarding the Python API, which is the primary interface for TRL.

Repositories

Other copies