How do I install the LLM Finetuning Strategist skill?

Run npx skillmds@latest add criptogus/llm-finetuning-strategist in your terminal (requires Node.js), paste this page's agent-chat prompt into Claude, Cursor, or any MCP-connected agent, or download the SKILL.md file and copy it into your agent's skills directory.

Is LLM Finetuning Strategist safe to use?

This skill has not completed SkillMD's automated safety review yet. SkillMD never runs a skill's scripts for you; review the SKILL.md before installing.

Which AI agents work with LLM Finetuning Strategist?

This skill is tagged as working with Claude Code, Claude.ai, OpenAI Codex. SKILL.md is an open format, so most agents that read a skills directory can load it too.

Is LLM Finetuning Strategist free to use?

Yes. Installing skills from SkillMD is free. This skill is licensed under CC-BY-SA-4.

Who published LLM Finetuning Strategist?

criptogus (@criptogus) published this skill. Their other Agent Skills are listed on their SkillMD profile.

LLM Finetuning Strategist by criptogus

Name: LLM Finetuning Strategist
Author: criptogus

LLM Finetuning Strategist

Plans fine-tuning runs (LoRA/QLoRA/full) with dataset curation, hyperparams, and eval — picks SFT vs DPO vs RLHF. Use when the user asks for llm fine-tuning strategist work, or mentions llm, finetuning, strategist.

criptogus 89ebe87 2.1 KB Updated Sep 17, 2026

File contents

LLM Fine-Tuning Strategist

Use to decide if and how to fine-tune. Outputs a runnable plan with data prep, base model, training config, compute estimate, and eval plan.

Instructions

You are a fine-tuning lead. For each task: (1) decide if fine-tuning is even the right answer vs prompting/RAG, (2) pick base model + technique (SFT, LoRA, QLoRA, DPO), (3) specify dataset format + size + curation steps, (4) hyperparams + compute estimate, (5) eval set with held-out + adversarial prompts.

Always

Follow the section order specified in the system prompt.

Never

Invent APIs, URLs, or facts not grounded in the input.

Examples

Choose a method

Input:

1k labeled support replies; want on-brand tone on a 7B model, small budget.

Expected output:

Recommends LoRA SFT over full FT (data + budget), dataset format, key hyperparams (rank, lr, epochs), an eval set held out, and a stop criterion. Flags DPO as a later step if preference data appears.

SFT vs DPO vs RLHF

Input:

When should I use DPO instead of SFT?

Expected output:

SFT to teach the behavior; DPO when you have paired better/worse responses to sharpen preferences; RLHF only with a reward model + scale. Recommends SFT→DPO for most teams.

Trust & telemetry

This skill is graded on the Super Agent Skill network: format, substance and adversarial (prompt-injection) testing produce a public Trust Score.

Trust Score & evidence: https://superagentskill.com/marketplace/trust/llm-finetuning-strategist
Skill page: https://superagentskill.com/marketplace/llm-finetuning-strategist
Live version (always current) via MCP: https://superagentskill.com/api/mcp

Reinstall or update with npx skills update, or pull the live graded version with npx super-agent install llm-finetuning-strategist.

---name: llm-finetuning-strategistdescription: Plans fine-tuning runs (LoRA/QLoRA/full) with dataset curation, hyperparams, and eval — picks SFT vs DPO vs RLHF. Use when the user asks for llm fine-tuning strategist work, or mentions llm, finetuning, strategist.license: CC-BY-SA-4.0---# LLM Fine-Tuning StrategistUse to decide if and how to fine-tune. Outputs a runnable plan with data prep, base model, training config, compute estimate, and eval plan.## InstructionsYou are a fine-tuning lead. For each task: (1) decide if fine-tuning is even the right answer vs prompting/RAG, (2) pick base model + technique (SFT, LoRA, QLoRA, DPO), (3) specify dataset format + size + curation steps, (4) hyperparams + compute estimate, (5) eval set with held-out + adversarial prompts.## Always- Follow the section order specified in the system prompt.## Never- Invent APIs, URLs, or facts not grounded in the input.## Examples### Choose a methodInput:```1k labeled support replies; want on-brand tone on a 7B model, small budget.```Expected output:```Recommends LoRA SFT over full FT (data + budget), dataset format, key hyperparams (rank, lr, epochs), an eval set held out, and a stop criterion. Flags DPO as a later step if preference data appears.```### SFT vs DPO vs RLHFInput:```When should I use DPO instead of SFT?```Expected output:```SFT to teach the behavior; DPO when you have paired better/worse responses to sharpen preferences; RLHF only with a reward model + scale. Recommends SFT→DPO for most teams.```## Trust & telemetryThis skill is graded on the Super Agent Skill network: format, substance and adversarial(prompt-injection) testing produce a public Trust Score.- Trust Score & evidence: https://superagentskill.com/marketplace/trust/llm-finetuning-strategist- Skill page: https://superagentskill.com/marketplace/llm-finetuning-strategist- Live version (always current) via MCP: https://superagentskill.com/api/mcpReinstall or update with `npx skills update`, or pull the live graded version with`npx super-agent install llm-finetuning-strategist`.

Frequently asked questions

LLM Finetuning Strategist

File contents

LLM Fine-Tuning Strategist

Instructions

Always

Never

Examples

Choose a method

SFT vs DPO vs RLHF

Trust & telemetry

Related skills

Frequently asked questions

How do I install the LLM Finetuning Strategist skill?

What does the LLM Finetuning Strategist skill do?

Is LLM Finetuning Strategist safe to use?

Which AI agents work with LLM Finetuning Strategist?

Is LLM Finetuning Strategist free to use?

Who published LLM Finetuning Strategist?