Back to catalog
Human reviewedOther

Skill dossier

fine-tuning-expert

Source declaration

Use when fine-tuning LLMs, training custom models, or adapting foundation models for specific tasks. Invoke for configuring LoRA/QLoRA adapters, preparing JSONL training datasets, setting hyperparameters for fine-tuning runs, adapter training, transfer learning, finetuning with Hugging Face PEFT, OpenAI fine-tuning, instruction tuning, RLHF, DPO, or quantizing and deploying fine-tuned models. Trigger terms include: LoRA, QLoRA, PEFT, finetuning, fine-tuning, adapter tuning, LLM training, model training, custom model.

The author's original description, preserved in English. It describes intent, not verified performance.

Use when fine-tuning LLMs, training custom models, or adapting foundation models for specific tasks. Invoke for configuring LoRA/QLoRA adapters, preparing JSONL training datasets, setting hyperparameters for fine-tuning runs, adapter training, transfer learning, finetuning with Hugging Face PEFT, OpenAI fine-tuning, instruction tuning, RLHF, DPO, or quantizing and deploying fine-tuned models. Trigger terms include: LoRA, QLoRA, PEFT, finetuning, fine-tuning, adapter tuning, LLM training, model training, custom model.

references

Research summary

Human reviewed

Skill Atlas editorial review · editorial-preview-1.0.0

A machine learning Skill for fine-tuning LLMs with LoRA and QLoRA, covering dataset validation, hyperparameters, evaluation against the base model and adapter merging for deployment.

Review scope

Deterministic static analysis of collected SKILL.md and supporting text/code. Machine-assisted candidates and findings await human editorial review; no dynamic behavior or effect is assessed.

Observable capabilities

  • Validates and formats JSONL training data before training starts.
  • Selects a PEFT method: LoRA for most tasks and QLoRA when GPU memory is constrained.
  • Configures Hugging Face PEFT and TRL training with warmup and monitors validation loss.
  • Evaluates against a held-out set and the base model, then merges the adapter and quantizes for deployment.

Limitations & scope

  • Needs GPU resources and model access; the example uses a Llama model ID that may require access approval.
  • It references a validate_dataset.py script that the skill file itself does not provide.
  • No training run or result quality has been verified here.

Relevant for

  • ML engineers adapting open models to specific tasks
  • Teams with constrained GPU memory using QLoRA

What ships with the skill

The file inventory comes from the collected source at the recorded commit. Paths and sizes are structural observations, not a measure of quality.

files
6
Markdown files
6
scripts
0
references
5
Size
84 KB
Structural complexity
Low

File inventory

PathTypeBytes
skills/fine-tuning-expert/references/dataset-preparation.mdReference16.1 KB
skills/fine-tuning-expert/references/deployment-optimization.mdReference16.7 KB
skills/fine-tuning-expert/references/evaluation-metrics.mdReference17.6 KB
skills/fine-tuning-expert/references/hyperparameter-tuning.mdReference16.2 KB
skills/fine-tuning-expert/references/lora-peft.mdReference10.8 KB
skills/fine-tuning-expert/SKILL.mdInstruction6.6 KB

SKILL.md frontmatter

{
  "name": "fine-tuning-expert",
  "description": "Use when fine-tuning LLMs, training custom models, or adapting foundation models for specific tasks. Invoke for configuring LoRA/QLoRA adapters, preparing JSONL training datasets, setting hyperparameters for fine-tuning runs, adapter training, transfer learning, finetuning with Hugging Face PEFT, OpenAI fine-tuning, instruction tuning, RLHF, DPO, or quantizing and deploying fine-tuned models. Trigger terms include: LoRA, QLoRA, PEFT, finetuning, fine-tuning, adapter tuning, LLM training, model training, custom model.",
  "license": "MIT",
  "metadata": {
    "author": "https://github.com/Jeffallan",
    "company": "https://synergetic.solutions",
    "version": "1.1.0",
    "domain": "data-ml",
    "triggers": "fine-tuning, fine tuning, finetuning, LoRA, QLoRA, PEFT, adapter tuning, transfer learning, model training, custom model, LLM training, instruction tuning, RLHF, model optimization, quantization",
    "role": "expert",
    "scope": "implementation",
    "output-format": "code",
    "related-skills": "devops-engineer"
  }
}

A reading of the operating sequence

This workflow reconstructs the apparent sequence from source material. Unless controlled evaluation is explicitly recorded, it is provisional—not an observed trace.

  1. 01
    Input

    Prepare the dataset

    Validate and format the data and run quality checks before training starts.

  2. 02
    Action

    Select method and train

    Choose LoRA or QLoRA based on GPU memory, then configure hyperparameters and monitor loss curves.

  3. 03
    Decision

    Check validation loss

    Validation loss must decrease; a plateau or increase signals overfitting.

  4. 04
    Output

    Evaluate and deploy

    Benchmark against the base model, then merge adapter weights, quantize and measure throughput before serving.

Potential risk surfaces

Static findings identify points to inspect before installation. They do not establish that a skill is safe, malicious, or effective.

Attention recommended

76 deterministic lexical finding(s) require review across the collected text. Findings can occur in documentation and do not prove execution or exploitability; binary-only files were not scanned.

Network accessObserved
File writesObserved
Credential useObserved
Note

File modification is described or implemented

64 occurrences

This line contains a possible file-write or redirection operation. Check destination paths and overwrite behavior; the static review did not write any file. This rule matched 64 location(s) in the collected text; 3 representative location(s) are cited. Matching text may be documentation, and the analyzer did not execute it.

High

File deletion is described or implemented

Deletion syntax occurs in this line, including possible documentation. Review its exact target, recursive scope, and safeguards. The analyzer did not execute it. This rule matched 1 location(s) in the collected text; 1 representative location(s) are cited. Matching text may be documentation, and the analyzer did not execute it.

Medium

Network access is mentioned

6 occurrences

A URL or network client appears in this source line. Review destinations, data sent, and trust boundaries; static text cannot establish actual network activity. This rule matched 6 location(s) in the collected text; 3 representative location(s) are cited. Matching text may be documentation, and the analyzer did not execute it.

High

Credential or secret material is referenced

This line resembles a credential literal or reads a sensitive environment value. Do not expose its contents; verify whether it is a placeholder, required secret, or accidentally committed material. This rule matched 1 location(s) in the collected text; 1 representative location(s) are cited. Matching text may be documentation, and the analyzer did not execute it.

Inspect all security findings (76 findings)
High

Dynamic command execution is referenced

4 occurrences

A shell, dynamic evaluation, or process execution primitive appears here. Inspect arguments and input control. Presence in source does not prove a command ran. This rule matched 4 location(s) in the collected text; 3 representative location(s) are cited. Matching text may be documentation, and the analyzer did not execute it.

Editorial security note

Check dataset provenance and licensing, and review any scripts before running them on GPU infrastructure or sensitive data.

Where context can be spent

Discovery, activation, and resource layers have different meanings. Exact values appear only when the official token-counting method was run with a named model.

Exact token counts unavailable

All token values, including per-file inventory, are null until the pipeline performs an official exact count with an explicit model.

No token measurement was recorded with the required official method. This dossier does not substitute character counts, file sizes, or estimates.

Controlled evaluation

No controlled evaluation

This skill has only the evidence identified above. No task success, token efficiency, latency, or quality advantage is claimed.

Not tested

Use with the source in view

Install command
git clone https://github.com/Jeffallan/claude-skills.git cd claude-skills # Copy this skill directory into one of: # ~/.claude/skills/fine-tuning-expert # .claude/skills/fine-tuning-expert
  1. Open the pinned source at https://github.com/Jeffallan/claude-skills/blob/67c1961b180dc92cf08983344627109e983e8f51/skills/fine-tuning-expert/SKILL.md.
  2. Review the license, prerequisites and all bundled reference files.
  3. Copy the skill directory into a personal or project-level Claude Code skills directory.
  4. Restart or refresh Claude Code so the skill can be discovered.
Verify installation

Ask Claude Code to draft a LoRA config and dataset validation plan, and confirm it includes evaluation before deployment.

Prerequisites

  • Needs GPU resources and model access; the example uses a Llama model ID that may require access approval.
  • It references a validate_dataset.py script that the skill file itself does not provide.
  • No training run or result quality has been verified here.

Example prompts

Representative task

Use the fine-tuning-expert skill to plan a QLoRA fine-tune of a 7B model on my instruction JSONL file: validate the dataset, propose LoraConfig and TrainingArguments, and define evaluation against the base model.

Inspect before use

Read the fine-tuning-expert skill at https://github.com/Jeffallan/claude-skills/blob/67c1961b180dc92cf08983344627109e983e8f51/skills/fine-tuning-expert/SKILL.md. Explain its workflow, required tools, permissions and the main risks before I decide whether to install it.

A decision, with its limits attached

Strengths

  • Defines checkpoints at each stage including data quality and loss curves.
  • Forbids deploying without held-out evaluation and a latency benchmark.

Considerations

  • Example hyperparameters are starting points and need tuning for each task.
  • Training and deployment steps can be costly in compute and time.
Research conclusion

A well-structured workflow for parameter-efficient fine-tuning that still relies on you to supply data, compute and judgment.

Traceable by construction

The dossier is a time-bound reading of a public artifact, not an endorsement. Use the source link to inspect the exact upstream material.

Jeffallanclaude-skillsskills/fine-tuning-expert
Source type
Community
Commit
67c1961b180dc92cf08983344627109e983e8f51
Collected
20 Nov 2026
Last checked
20 Nov 2026
Upstream license
MIT
Repository stars
12,464
Content basis

Deterministic lexical parsing, YAML frontmatter validation, file inventory, source-described workflow extraction, and heuristic provisional classification of the supplied files only. Generated by machine; no LLM, execution, dynamic testing, or human editorial approval.