Axolotl — тонкая настройка LLM

★ 7.7 · ai-tooling

axolotl is a Claude Code skill that provides expert guidance for fine-tuning large language models with the Axolotl framework. It covers 100+ models and all major training methods — LoRA, QLoRA, DPO, KTO, ORPO, GRPO — as well as multimodal architectures. Training runs are configured via YAML files; the skill helps set up distributed training with FSDP and context parallelism, compress checkpoints to save up to 40% disk space, and identify bottlenecks using NCCL tests. Reference files on the API, dataset formats, and common code patterns are extracted from official Axolotl documentation and bundled with the skill. It is aimed at ML engineers who want to quickly launch fine-tuning on custom data or debug an existing training pipeline.