claudegoodies
Skill

axolotl

From NousResearch

Axolotl: YAML LLM fine-tuning (LoRA, DPO, GRPO).

Provides reference docs and config patterns for fine-tuning LLMs with Axolotl via YAML (LoRA, QLoRA, DPO, KTO, ORPO, GRPO).

Use it when

  • Writing or debugging Axolotl YAML training configs
  • Setting up FSDP, context parallelism, or multi-GPU training
  • Choosing dataset formats or writing custom prompt strategies
  • Configuring compressed model saving or NCCL bandwidth testing

Skip it if

  • Not using Axolotl specifically for fine-tuning
  • Needs Linux or macOS with GPU/DeepSpeed setup already in place
  • Auto-generated reference docs, not a working tool or script

Facts

Repository
NousResearch/hermes-agent
Status
Actively maintained
Last commit
Source file
optional-skills/mlops/training/axolotl/SKILL.md

Source preview

The instructions Claude Code reads when this skill runs.

# Axolotl Skill

## What's inside

Expert guidance for fine-tuning LLMs with Axolotl — YAML configs, 100+ models, LoRA/QLoRA, DPO/KTO/ORPO/GRPO, multimodal support.

Comprehensive assistance with axolotl development, generated from official documentation.

## When to Use This Skill

This skill should be triggered when:
- Working with axolotl
- Asking about axolotl features or APIs
- Implementing axolotl solutions
- Debugging axolotl code
- Learning axolotl best practices

## Quick Reference

### Common Patterns

**Pattern 1:** To validate that acceptable data transfer speeds exist for your training job, running NCCL Tests can help pinpoint bottlenecks, for example:

```
./build/all_reduce_perf -b 8 -e 128M -f 2 -g 3
```

**Pattern 2:** Configure your model to use FSDP in the Axolotl yaml. For example:

```
fsdp_version: 2
fsdp_config:
  offload_params: true
  state_dict_type: FULL_STATE_DICT
  auto_wrap_policy: TRANSFORMER_BASED_WRAP
  transformer_layer_cls_to_wrap: LlamaDecoderLayer
  reshard_after_forward: true
```

**Pattern 3:** The context_parallel_size should be a divisor of the total number of GPUs. For example:

```
context_parallel_size
```

**Pattern 4:** For example: - With 8 GPUs and no sequence parallelism: 8 different batches processed per step - With 8 GPUs and context_parallel_size=4: Only 2 different batches processed per step (each split across 4 GPUs) - If yo
View full source on GitHub →

Other skills