axolotl

Fine-tune LLMs with YAML configs by installing Axolotl, creating a config file for your model, and starting training with your dataset. This skill provides official references and patterns for LoRA/QLoRA optimizations to accelerate model customization.

Yes, DPO and GRPO are supported for RLHF workflows. This skill provides expert guidance for integrating these multiple RLHF strategies, alongside ORPO, into your end-to-end training workflows using Axolotl.

Fine-tune LLMs for multimodal data using Axolotl's supported multimodal data pipelines. This skill offers comprehensive guidance and practical tutorials to help you build end-to-end training and evaluation workflows for multimodal models.

Yes, this skill works with DeepSpeed and accelerate, which are listed dependencies. It provides expert guidance for Axolotl-based training workflows that utilize these frameworks for distributed training and hardware acceleration.

Use QLoRA over LoRA when you need optimized memory efficiency for model customization. This skill provides official references and example patterns for both LoRA and QLoRA optimizations to help you accelerate development.

Yes, PyTorch and transformers are required dependencies to start training. You need these foundational libraries installed alongside Axolotl to execute the YAML-configured training workflows and model evaluation.