axolotl

Configuring an Axolotl YAML file for LoRA fine-tuning involves mapping specific configuration keys to parameter-efficient training scenarios, defining target modules and rank within the YAML structure. This Skill validates these configurations against official Axolotl documentation to ensure accurate setup.

Setting up DPO or GRPO training with Axolotl requires accurately mapping Axolotl configuration keys to supported preference and reinforcement learning objectives. This Skill translates official guidance into practical YAML configuration patterns, reducing trial-and-error for these specific training paradigms.

Yes, you can use FSDP and context parallelism for distributed training in Axolotl by setting the appropriate distributed training options in your YAML config. This Skill provides examples and constraints for configuring these distributed knobs effectively.

Yes, Axolotl supports saving and compressing checkpoints for downstream inference through specific YAML configuration options. This Skill guides you in applying the correct settings to handle checkpoint compression and prepare models for inference deployment.

Handling dataset formats and schemas when using Axolotl for fine-tuning requires matching your dataset schema to Axolotl's expected formats. This Skill provides practical troubleshooting-style guidance to manage common dataset schema expectations and tokenizer configurations.