LLM Finetuning

Rewire a pretrained model to your task.

A hands-on tour of finetuning: instruction tuning by hand, then Hugging Face, Unsloth, LoRA, and QLoRA at scale. Closes with research-grade projects on subliminal learning and retrieval-augmented finetuning (RAFT).

intermediatellmfinetuninglorapeft

40 capsules144 figures~8 hoursby Dr. Raj Dandekar

Read on your Kindle or e-reader

Download the EPUB and read offline — perfect for the train. Works on Kindle, Kobo, Apple Books & more.

Send to Kindle →

00Why Finetune at All5 capsules

Why Finetune at All — 5 chapters.

01What Finetuning Actually Isconceptfree10 min 02Pretraining vs Finetuningconceptfree10 min 03Knowledge vs Behaviorintuition🔒12 min 04Transfer Learning and Intrinsic Dimensionalitymath🔒10 min 05Who Finetuning Is Forconcept🔒12 min

01Finetuning vs RAG6 capsules

Finetuning vs RAG — 6 chapters.

06The Central Tradeoffconcept🔒11 min 07How RAG Worksconcept🔒11 min 08Structured vs Unstructured RAGconcept🔒10 min 09Case Study: A Financial Chatbotproject🔒10 min 10The Technical Limits of RAGdeep-dive🔒12 min 11Few-Shot Prompting vs Finetuningintuition🔒12 min

02Instruction Finetuning from Scratch6 capsules

Instruction Finetuning from Scratch — 6 chapters.

12Anatomy of an Instruction Datasetcode🔒11 min 13Tokenizing and Formatting the Datacode🔒12 min 14The Finetuning Training Loopcode🔒12 min 15Sparse Finetuning and Layer Normsdeep-dive🔒10 min 16Evaluating Your Finetuned Modelconcept🔒14 min 17How OpenAI Uses Finetuningconcept🔒11 min

03Finetuning with Hugging Face & Unsloth7 capsules

Finetuning with Hugging Face & Unsloth — 7 chapters.

18The Hugging Face Ecosystemconcept🔒12 min 19Loading and Testing the Tiny Stories Modelcode🔒12 min 20Instruction vs Classification Finetuningconcept🔒11 min 21Adding a Custom Classification Headcode🔒11 min 22Data Collation and Dynamic Paddingcode🔒10 min 23Left vs Right Paddingdeep-dive🔒11 min 24A Finetuning Project with Unslothproject🔒12 min

04Parameter-Efficient Finetuning (PEFT)7 capsules

Parameter-Efficient Finetuning (PEFT) — 7 chapters.

25Why Full Finetuning Hurtsconcept🔒11 min 26The LoRA Ideamath🔒11 min 27The Math Behind LoRAmath🔒13 min 28Implementing LoRA in GPT-2code🔒13 min 29The PEFT Training Loopcode🔒14 min 30Tuning LoRA Hyperparametersdeep-dive🔒13 min 31QLoRA and Quantized Finetuningconcept🔒13 min

05Research Projects6 capsules

Research Projects — 6 chapters.

32Replicating a Research Paperproject🔒11 min 33Subliminal Learning Explaineddeep-dive🔒11 min 34Generating Training Data at Scalecode🔒11 min 35Finetuning on the OpenAI Platformproject🔒12 min 36RAFT: Retrieval-Augmented Finetuningdeep-dive🔒10 min 37Setting Up a RAFT Projectproject🔒12 min

06Putting It Together3 capsules

Putting It Together — 3 chapters.

38Choosing Your Finetuning Strategyconcept🔒12 min 39The Workshop Code and Resourcescode🔒12 min 40Capstone: Ship a Finetuned Modelproject🔒12 min

Ratings & reviews

1,607 readers

4.5

445 ratings

5 ★53%

4 ★27%

3 ★10%

2 ★5%

1 ★4%

Rate this book

Jonas Bergström

last week

A few sections moved fast for me, but re-reading with the figures open made everything land. Highly recommend.

Kevin Park

4 weeks ago

This is the resource I wish I had when I started. Clear mental models, zero fluff.

Aarav Mehta

last month

Concise capsules I can finish over coffee, yet each one taught me something I actually use.