Learn LLM principles and deployment using free Colab notebooks
Follow Colab notebooks that run directly in the browser to manually handle the entire process, from tokenizing text into numbers to fine-tuning models and final deployment. This course starts with mathematical foundations and covers the latest LLM architectures and alignment techniques, serving as free educational material that combines theory with practice. You can start learning immediately with a free Google Colab account without complex environment setup.
A learning path divided into three stages
The course is structured into three parts based on the learner's background and goals. The first, 'LLM Fundamentals', is optional and covers the mathematics, Python, and neural network basics required for machine learning. The second, 'LLM Scientist', focuses on building the best LLMs using the latest techniques. The final 'LLM Engineer' section concentrates on the practical work of creating and deploying LLM-based applications. Each section can be studied independently, and you can refer to only the parts you need.
Hands-on practice with Colab notebooks
The core of this course is Google Colab notebooks that run directly in the browser.
- Fine-tuning practice Learn step-by-step how to fine-tune Llama and Mistral models using various techniques such as Unsloth, ORPO, DPO, and QLoRA.
- Model optimization Cover the process of quantizing models into various formats like GGUF, GPTQ, and EXL2 to enable execution on consumer hardware.
- Data preparation Practice building datasets necessary for post-training, including synthetic data generation, deduplication, and quality filtering.
Distinguishing roles from scientist to engineer
This course clearly distinguishes between two main perspectives in LLM development. The 'LLM Scientist' section deeply covers the theory and technology for improving the model itself, including transformer architecture, pre-training, preference alignment (DPO, PPO, etc.), and evaluation methodologies. In contrast, the 'LLM Engineer' section focuses on the practical work of applying these models to real-world applications and deploying them. Readers can selectively study the internal structure of models or the technologies needed to create the final user experience, depending on their interests.
Prerequisites and limitations before learning
Although this course is free, some advanced practices may require GPU resources. The 'LLM Fundamentals' section is optional, but it is recommended that readers unfamiliar with mathematics or Python refer to this part first. Since all notebooks run on Google Colab, a Google account is required. Additionally, while the course is structured around concepts that are not dependent on specific frameworks or languages, Python and PyTorch are used in the practice code. Readers can selectively skip or repeat sections based on their background knowledge.
