Hugging Face Agents Course MOC
This map provides a theoretical and practical traversal of the Hugging Face Agents Course, focusing on the core mechanics of reasoning, tool-use, and production orchestration for agent builders. It assumes baseline familiarity with LLM concepts and Python-style tooling.
📚 Course Literature
- lit-hf-agents-fundamentals: Unit 1 (Fundamentals) & Unit 2 (Frameworks).
- lit-hf-agents-applications: Unit 3 (Use Cases) & Unit 4 (Evaluation).
- lit-hf-agents-bonus: Bonus Material (Fine-tuning, Observability, Gaming).
Fundamentals (Unit 1)
- agent-thought-cycle: Thought-Action-Observation.
- react-pattern: Reasoning + Acting.
- chat-templates: Bridging roles and special tokens.
- agent-tools: Defining the "Body" of an agent through available capabilities.
- agent-actions-unit: Actions, code agents, and environment engagement during orchestration.
Frameworks & Toolkits (Unit 2)
- smolagents: Freedom and Code Agents (HF).
- llamaindex: Data Agency and RAG Workflows.
- langgraph: Control and Graph-Based Orchestration.
Applications (Unit 3 & Bonus 3)
- gala-agent-use-case: Building Alfred the Gala Host.
- agentic-rag: Implementation patterns for dynamic retrieval.
- agents-in-games: Autonomous NPCs and strategy.
- pokemon-battle-agent: Case study in turn-based interaction.
Advanced Theory (Bonus 1 & 2)
- function-calling: Native model capabilities.
- lora: Efficient fine-tuning for agentic tasks.
- agent-observability: Traces, Spans, and OpenTelemetry.
- agent-evaluation: Measuring quality and helpfulness.
- llm-as-a-judge: Automated grading.
The Benchmark (Unit 4)
- gaia-benchmark: A prominent benchmark for general AI assistants.