KoreaDevKNOWLEDGE SHARING

LLM EDUCATION · 8 INTEGRATED COURSES

How to choose, connect, and operate a local LLM safely

The 20 core units keep their full depth, and content that is best read together is grouped into 8 extended courses. Choose a course on the left to follow its textbook, diagrams, labs, and assessments on a single page.

01 Start the integrated courseHardware for your environment
My learning progress0 / 8 courses
0%Complete it once, at the end of each integrated course.
User courses
8
Verified core units
20
Training method
Long reading + hands-on lab

LLM EDUCATION

8 integrated courses

  1. 01LLM Foundations

    LLM principles and language

    Explains the full flow from input through tokens to a response without memorizing unfamiliar terms separately.

    Absolute beginner
  2. 02Memory & Quantization

    Memory and quantization

    Judge the differences among model file size, runtime memory, context length, and active parameters with real numbers.

    Foundations
  3. 03AI Hardware Selection

    Accelerators and system selection

    Compare workload, memory, software ecosystem, and operating conditions before product names.

    Foundations
  4. 04Models & Runtimes

    Models and runtimes

    Find and download the model, reproduce it with a compatible runtime, and record the source and version as evidence.

    Foundations
  5. 05Prompt & Agent Workflows

    Prompt and agent workflows

    Do not trust generated results as-is; design tool permissions, failure paths, and human review points.

    Practical
  6. 06Data & Fine-tuning

    Data and fine-tuning

    Fix the data contract and baseline first, then compare the effect and cost of small adapter training.

    Practical
  7. 07RAG & Evaluation

    RAG and evaluation

    Retrieve documents to link evidence to answers, and revalidate quality before and after changes with the same question set.

    Practical
  8. 08Secure Serving

    Secure serving

    Before exposing the model API, combine security boundaries, observability metrics, rollback, and user outcomes into a single acceptance criterion.

    Capstone