What models does it recommend?

It provides data-driven recommendations for popular models like Qwen for Chinese tasks, DeepSeek for reasoning, Llama-3 for ecosystem support, and Phi-4 for lightweight deployment.

Can it help if I have very little training data?

Yes, it includes a decision tree for small datasets, suggesting few-shot prompting for under 100 samples or specific LoRA configurations for datasets between 100 and 500 samples.

Does this skill perform the actual training?

It acts as a guide and architect, helping you define the configuration, success criteria, and strategy, which you can then execute using the generated task.yaml.

When should I use the llm-coach skill?

Use it whenever you are planning to train, fine-tune, or build an NLP model and need guidance on the best approach, model selection, or success metrics.

LLM Training & Fine-Tuning Coach

Name: LLM Training & Fine-Tuning Coach
Author: p988744

byp988744

データサイエンスとML

Guides users through the complete lifecycle of LLM fine-tuning, from initial goal setting to selecting optimal training strategies and models.

概要

The llm-coach skill acts as a specialized consultant for Natural Language Processing tasks, helping users navigate the complexities of model training. It utilizes a structured four-stage coaching process—exploration, goal clarification, strategy recommendation, and execution—to ensure users select the right base models (such as Qwen, DeepSeek, or Llama) and techniques (SFT, LoRA, DPO) based on their specific data constraints, hardware resources, and performance targets.

主な機能

Comprehensive version management strategies including Semantic and Hybrid naming
Standardized YAML task configuration generation for consistent training runs
Base model selection guidance for specific languages and reasoning tasks
Structured pain point exploration and business context discovery
0 GitHub stars
Automated recommendation engine for training methods like LoRA, SFT, and DPO

ユースケース

Establishing success criteria and evaluation baselines for new ML projects
Determining the most cost-effective training strategy based on available GPU memory
Clarifying vague NLP requirements such as 'I want to do sentiment analysis'

概要

主な機能

Comprehensive version management strategies including Semantic and Hybrid naming
Standardized YAML task configuration generation for consistent training runs
Base model selection guidance for specific languages and reasoning tasks
Structured pain point exploration and business context discovery
0 GitHub stars
Automated recommendation engine for training methods like LoRA, SFT, and DPO

ユースケース

Establishing success criteria and evaluation baselines for new ML projects
Determining the most cost-effective training strategy based on available GPU memory
Clarifying vague NLP requirements such as 'I want to do sentiment analysis'