Toolverse
All skills

peft-fine-tuning

by davila7

Parameter-efficient fine-tuning for LLMs using LoRA, QLoRA, and 25+ methods. Use when fine-tuning large models (7B-70B) with limited GPU memory, when you need to train u003c1% of parameters with minimal accuracy loss, or for multi-adapter serving. HuggingFace's official library

Installation

Pick a client and clone the repository into its skills directory.

Installation

Quick info

Author
davila7
Category
Security
Views
1

About this skill

Parameter-efficient fine-tuning for LLMs using LoRA, QLoRA, and 25+ methods. Use when fine-tuning large models (7B-70B) with limited GPU memory, when you need to train u003c1% of parameters with minimal accuracy loss, or for multi-adapter serving. HuggingFace's official library integrated with transformers ecosystem.

How to use

  1. Zainstaluj bibliotekę PEFT wraz z zależnościami: uruchom pip install peft transformers accelerate bitsandbytes datasets. Opcjonalnie dodaj bitsandbytes dla wsparcia kwantyzacji, jeśli planujesz pracować z modelami 70B na GPU z 24GB pamięci.

  2. Załaduj bazowy model i tokenizer z HuggingFace Hub, na przykład Llama 3.1 8B lub inny model obsługiwany przez bibliotekę transformers.

  3. Skonfiguruj LoRA poprzez LoraConfig, określając parametry takie jak ranga (rank), współczynnik skalowania (lora_alpha) i warstwy docelowe. Dla większości przypadków domyślne wartości działają dobrze.

  4. Owinięcie modelu w get_peft_model() aktywuje dostrajanie efektywne — model będzie trenować tylko adaptery, a nie wszystkie wagi.

  5. Przygotuj swój zbiór danych (własny lub z biblioteki datasets) i uruchom trening za pomocą Trainer z TrainingArguments. Określ liczbę epok, rozmiar batcha i ścieżkę zapisu.

  6. Po treningu zapisz adapter (model.save_pretrained()) — zajmie on kilka MB. Możesz załadować go później i łączyć z bazowym modelem do wnioskowania lub dalszego dostrajania dla innych zadań.

Related skills

architect-review

by sickn33

Master software architect specializing in modern architecture patterns, clean architecture, microservices, event-driven systems, and DDD. Reviews system designs and code changes for architectural integrity, scalability, and maintainability. Use PROACTIVELY for architectural

Security
2773

skill-writer

by pytorch

Guide users through creating Agent Skills for Claude Code. Use when the user wants to create, write, author, or design a new Skill, or needs help with SKILL.md files, frontmatter, or skill structure.

Security
15116

reverse-engineering-tools

by gmh5225

Guide for reverse engineering tools and techniques used in game security research. Use this skill when working with debuggers, disassemblers, memory analysis tools, binary analysis, or decompilers for game security research.

Security
3168

google-analytics

by davila7

Analyze Google Analytics data, review website performance metrics, identify traffic patterns, and suggest data-driven improvements. Use when the user asks about analytics, website metrics, traffic analysis, conversion rates, user behavior, or performance optimization.

Security
1260

youtube-watcher

by openclaw

Fetch and read transcripts from YouTube videos. Use when you need to summarize a video, answer questions about its content, or extract information from it.

Security
2231

backend-security-coder

by sickn33

Expert in secure backend coding practices specializing in input validation, authentication, and API security. Use PROACTIVELY for backend security implementations or security code reviews.

Security
1133