跳转到主要内容
此内容尚未提供您的语言版本,正在以英文显示。

PEFT Fine Tuning

技能 已验证 活跃

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 <1% of parameters with minimal accuracy loss, or for multi-adapter serving. HuggingFace's official library integrated with transformers ecosystem.

目的

To enable users to fine-tune large language models efficiently on limited hardware by training only a small fraction of model parameters.

功能

  • Parameter-efficient fine-tuning (PEFT) with LoRA, QLoRA, and 25+ methods
  • Guidance for fine-tuning 7B-70B models on consumer GPUs
  • Code examples for standard LoRA and memory-efficient QLoRA
  • Detailed instructions on loading, merging, and managing adapters
  • Performance benchmarks and troubleshooting for common issues

使用场景

  • Fine-tuning large LLMs (7B-70B) with limited GPU memory
  • Training less than 1% of model parameters for minimal accuracy loss
  • Enabling multi-adapter serving from a single base model
  • Rapid iteration on task-specific adapters for LLMs

非目标

  • Full fine-tuning for maximum quality when compute budget is not a constraint
  • Training models smaller than 1B parameters
  • Scenarios requiring updating all model weights due to significant domain shift

工作流

  1. Install necessary libraries (PEFT, Transformers, PyTorch, etc.)
  2. Configure LoRA or QLoRA parameters (rank, alpha, target modules)
  3. Load base model and apply PEFT configuration
  4. Prepare dataset and tokenize data
  5. Train the model using provided training arguments
  6. Save and optionally merge the trained adapter

实践

  • Fine-tuning methodology
  • LLM optimization
  • Adapter management

先决条件

  • Python 3.8+
  • pip install peft transformers torch bitsandbytes datasets accelerate

安装

npx skills add davila7/claude-code-templates

通过 npx 运行 Vercel skills CLI(skills.sh)— 需要本地安装 Node.js,以及至少一个兼容 skills 的智能体(Claude Code、Cursor、Codex 等)。前提是仓库遵循 agentskills.io 格式。

质量评分

已验证
96 /100
1 day ago 分析

信任信号

最近提交1 day ago
星标27.2k
许可证MIT
状态
查看源代码

类似扩展

Peft Fine Tuning

99

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 <1% of parameters with minimal accuracy loss, or for multi-adapter serving. HuggingFace's official library integrated with transformers ecosystem.

技能
Orchestra-Research

Fine Tuning Expert

98

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.

技能
jeffallan

Unsloth

98

Expert guidance for fast fine-tuning with Unsloth - 2-5x faster training, 50-80% less memory, LoRA/QLoRA optimization

技能
Orchestra-Research

Implementing Llms Litgpt

98

Implements and trains LLMs using Lightning AI's LitGPT with 20+ pretrained architectures (Llama, Gemma, Phi, Qwen, Mistral). Use when need clean model implementations, educational understanding of architectures, or production fine-tuning with LoRA/QLoRA. Single-file implementations, no abstraction layers.

技能
Orchestra-Research

Implementing Llms Litgpt

100

Implements and trains LLMs using Lightning AI's LitGPT with 20+ pretrained architectures (Llama, Gemma, Phi, Qwen, Mistral). Use when need clean model implementations, educational understanding of architectures, or production fine-tuning with LoRA/QLoRA. Single-file implementations, no abstraction layers.

技能
davila7

Unsloth

100

Expert guidance for fast fine-tuning with Unsloth - 2-5x faster training, 50-80% less memory, LoRA/QLoRA optimization

技能
davila7