Automate LLM fine-tuning configuration and execution via YAML files.
npx skills add https://github.com/DarkArty07/Aether-Agents --skill axolotl-darkarty07
Please help me install this Agent Skill. Skill: axolotl Source: https://github.com/DarkArty07/Aether-Agents/tree/main/home/skills/mlops/training/axolotl Command: npx skills add https://github.com/DarkArty07/Aether-Agents --skill axolotl-darkarty07
Manually configuring Axolotl for LLM fine-tuning is complex and error-prone, requiring deep familiarity with YAML schemas, distributed training setups, and dataset formatting rules for diverse fine-tuning paradigms.
Use the axolotl skill to create a valid YAML configuration for 4-bit QLoRA fine-tuning of a Mistral 7B model with DeepSpeed ZeRO-3 and a custom instruction dataset.
High-intent search queries and answers about installing and using this skill.
To configure LLM fine-tuning with LoRA and DeepSpeed, you use the Axolotl framework to generate valid YAML configuration files. This skill automates setup for 4-bit QLoRA, DeepSpeed ZeRO-3, and distributed training across 100+ model architectures.
Yes, Axolotl supports preference-based post-training including DPO, GRPO, ORPO, and KTO. You can configure these RLHF workflows alongside supervised fine-tuning and multimodal training by defining the paradigm in your YAML configuration.
The best way to format datasets for supervised fine-tuning is to use Axolotl's supported formats, which include pre-tokenized, template-free, conversation, and instruction datasets. This skill provides formatting assistance and troubleshooting for chat template and tokenization errors.
You set up distributed training with FSDP and mixed precision by configuring Axolotl YAML files. This skill provides guidance for FSDP, DeepSpeed, context parallelism, and mixed precision (FP16/BF16/FP8) to optimize performance and memory usage.
NCCL communication bottlenecks during distributed training can arise from suboptimal network configurations. This skill enables training performance optimization by validating NCCL communication speeds and debugging EOS token masking issues.
Supporter09Compress large language models into 4-bit weights with activation-aware quantization.
ScientiaCapitalExport fine-tuned models to GGUF and deploy with vLLM or Ollama.
BagelHoleSet up QLoRA, LoRA, and full fine-tuning infrastructure with Hugging Face TRL and Axolotl.
jasrajtulsiLoads and fine-tunes Hugging Face Transformers models for AI workflows.
