此為 https://aihao.tw/llm 課程的 Example code

Setup

請先安裝 uv https://github.com/astral-sh/uv

uv sync

source .venv/bin/activate

複製 .env.example 至 .env 然後分別加上 API Key

Prompt Engineering Example (Braintrust)

https://www.braintrust.dev/

python eval_braintrust_1_dataset.py
python eval_braintrust_2_category.py
python eval_braintrust_2_recommend.py

Prompt Engineering Example (LangSmith)

https://www.langchain.com/langsmith/

python eval_langsmith_1_dataset.py
python eval_langsmith_2_category.py
python eval_langsmith_2_recommend.py
python eval_langsmith_2_recommend_openevals.py

Prompt Engineering Example (Promptfoo)

安裝 https://www.promptfoo.dev/

npx promptfoo@latest eval -c promptfooconfig.yaml
npx promptfoo@latest view

FastAPI SSE Example

Copy .env.example to .env and edit it

uvicorn main:app --reload

Open

http://localhost:8000/static/completion.html
http://localhost:8000/static/structured_output.html
http://localhost:8000/static/agent.html

Build Your Own Code Interpreter

參考自 https://cookbook.openai.com/examples/object_oriented_agentic_approach/secure_code_interpreter_tool_for_llm_agents

docker build -t python_sandbox:latest .

docker run -d --name sandbox --network none --cap-drop all --pids-limit 64 --tmpfs /tmp:rw,size=64M python_sandbox:latest sleep infinity

python demo_code_interpreter.py

MCP demo

uv add "mcp[cli]"

mcp install mcp install my_mcp_server.py 這會安裝到 claude app 中 mcp dev my_mcp_server.py 這會打開 MCP Inspector 在 http://127.0.0.1:6274