菜单
Expert guidance for fine-tuning LLMs with Axolotl - YAML configs, 100+ models, LoRA/QLoRA, DPO/KTO/ORPO/GRPO, multimodal support
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Run LLM post-training on Tinker with CPU-side orchestration and remote GPU execution. Use when preparing or launching SFT, DPO, or PPO-style runs, checkpointing, sampling checkpoints, or resuming long-running jobs through Hermes Research Agent.
Run durable end-to-end LLM post-training research loops from zero-spec ideas to finished reports. Use when the goal is autonomous literature review, hypothesis selection, experiment planning, Tinker training, evaluation, checkpointing, and resumable long-running research work.
Plan and interpret model evaluations and ablations for post-training research. Use when comparing checkpoints, designing ablations, selecting benchmarks, or summarizing what changed after training.
Convert literature findings into concrete LLM experiment plans. Use when the agent has papers, blog posts, or benchmark notes and needs to turn them into hypotheses, training plans, dataset choices, and evaluation criteria.
Generate and score candidate LLM research ideas from sparse user goals. Use when a project starts with no concrete benchmark, dataset, or training recipe and the agent needs to propose plausible, testable directions.
Produce clear iteration memos and final research summaries from project artifacts. Use when converting experiment state, training outcomes, evaluation results, and open questions into reports for later review.
| name | axolotl |
| description | Expert guidance for fine-tuning LLMs with Axolotl - YAML configs, 100+ models, LoRA/QLoRA, DPO/KTO/ORPO/GRPO, multimodal support |
| version | 1.0.0 |
| author | Orchestra Research |
| license | MIT |
| dependencies | ["axolotl","torch","transformers","datasets","peft","accelerate","deepspeed"] |
| metadata | {"hermes":{"tags":["Fine-Tuning","Axolotl","LLM","LoRA","QLoRA","DPO","KTO","ORPO","GRPO","YAML","HuggingFace","DeepSpeed","Multimodal"]}} |
Comprehensive assistance with axolotl development, generated from official documentation.
This skill should be triggered when:
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 your per-GPU micro_batch_size is 2, the global batch size decreases from 16 to 4
context_parallel_size=4
Pattern 5: Setting save_compressed: true in your configuration enables saving models in a compressed format, which: - Reduces disk space usage by approximately 40% - Maintains compatibility with vLLM for accelerated inference - Maintains compatibility with llmcompressor for further optimization (example: quantization)
save_compressed: true
Pattern 6: Note It is not necessary to place your integration in the integrations folder. It can be in any location, so long as it’s installed in a package in your python env. See this repo for an example: https://github.com/axolotl-ai-cloud/diff-transformer
integrations
Pattern 7: Handle both single-example and batched data. - single example: sample[‘input_ids’] is a list[int] - batched data: sample[‘input_ids’] is a list[list[int]]
utils.trainer.drop_long_seq(sample, sequence_len=2048, min_sequence_len=2)
Example 1 (python):
cli.cloud.modal_.ModalCloud(config, app=None)
Example 2 (python):
cli.cloud.modal_.run_cmd(cmd, run_folder, volumes=None)
Example 3 (python):
core.trainers.base.AxolotlTrainer(
*_args,
bench_data_collator=None,
eval_data_collator=None,
dataset_tags=None,
**kwargs,
)
Example 4 (python):
core.trainers.base.AxolotlTrainer.log(logs, start_time=None)
Example 5 (python):
prompt_strategies.input_output.RawInputOutputPrompter()
This skill includes comprehensive documentation in references/:
Use view to read specific reference files when detailed information is needed.
Start with the getting_started or tutorials reference files for foundational concepts.
Use the appropriate category reference file (api, guides, etc.) for detailed information.
The quick reference section above contains common patterns extracted from the official docs.
Organized documentation extracted from official sources. These files contain:
Add helper scripts here for common automation tasks.
Add templates, boilerplate, or example projects here.
To refresh this skill with updated documentation: