LLM-Cookbook

A collection of ready-to-run notebooks and practical guides for training and fine-tuning large language models (LLMs) in 2025.

---

Core Techniques Covered

Supervised Fine-Tuning (SFT)

Train models on labeled input-output pairs using Teacher Forcing.

Direct Preference Optimization (DPO)

Align models with human preferences using pairwise comparisons.

Quantization

Reduce memory and compute requirements using 8-bit and 4-bit model formats via BitsAndBytes.

Parameter-Efficient Fine-Tuning (PEFT) with LoRA

Fine-tune models efficiently by injecting low-rank adapters, using the peft library.

Signal-to-Noise Ratio (SNR) Analysis

Understand layer-wise learning dynamics with the spectrum library.

Evaluation Metrics

Assess performance with metrics like Accuracy, BLEU, ROUGE, BERTScore, and Levenshtein distance.

---

Dataset Creation

Dataset-Creation-DPO.ipynb

Dataset-Creation-SFT.ipynb

---

Training

FineTune-LLM-using-DPO.ipynb

FineTune-LLM-using-SFT.ipynb

⚠️ These notebooks are optimized for demonstration and may use small datasets and short training times.