دوره فاین‌تیونینگ و بهینه‌سازی مدل‌ها

چه چیزی یاد خواهید گرفت؟

فصل 1: انواع Fine-tuning

Full Fine-tuning vs Parameter-Efficient Fine-tuning (PEFT)

Supervised Fine-tuning (SFT)

Instruction Fine-tuning

Domain-specific Fine-tuning

Constitutional AI Training

Dataset Preparation

پروژه عملی: Fine-tuning برای دامنه خاص

فصل 2: PEFT و تکنیک‌های بهینه‌سازی

Low-Rank Adaptation (LoRA)

Quantized LoRA (QLoRA)

AdaLoRA و IA3 Techniques

Adapters و Prefix Tuning

4-bit Normal Float (NF4)

Performance vs Quality Trade-offs

پروژه عملی: LoRA Implementation

فصل 3: Reinforcement Learning from Human Feedback (RLHF)

اصول RLHF

Reward Model Training

Proximal Policy Optimization (PPO)

Direct Preference Optimization (DPO)

Human Feedback Collection

RLAIF (AI Feedback)

پروژه عملی: RLHF Pipeline

فصل 4: Model Optimization Techniques

Quantization: Post-training و Quantization-Aware Training

GPTQ و AWQ Quantization

Pruning: Structured و Unstructured

Knowledge Distillation

Model Compression Strategies

Edge Deployment Optimization

پروژه عملی: Model Compression Pipeline

فصل 5: Local Model Deployment

Ollama Setup و Management

vLLM - High-Performance Inference

Text Generation Inference (TGI)

LM Studio - GUI Model Management

Llamafile - Single-File Deployment

Model Quantization for Local Use

پروژه عملی: Local LLM Server