工作坊示例 - 快速参考卡

最后更新：2025年10月8日

快速开始

# 1. Ensure Foundry Local is running
foundry service status
foundry model run phi-4-mini

# 2. Install dependencies
pip install -r Workshop/requirements.txt

# 3. Run a sample
cd Workshop/samples
python -m session01.chat_bootstrap "What is edge AI?"
​

示例概览

️ 环境变量

必需

# Choose model
set FOUNDRY_LOCAL_ALIAS=phi-4-mini

# Override endpoint (optional)
set FOUNDRY_LOCAL_ENDPOINT=http://localhost:8000

# Show token usage
set SHOW_USAGE=1
​

会话特定

# Session 02: RAG
set RAG_QUESTION="What is local inference?"
set EMBED_MODEL=sentence-transformers/all-MiniLM-L6-v2

# Session 03: Benchmarking
set BENCH_MODELS=phi-4-mini,qwen2.5-0.5b
set BENCH_ROUNDS=3
set BENCH_STREAM=1

# Session 04: Comparison
set SLM_ALIAS=phi-4-mini
set LLM_ALIAS=qwen2.5-7b

# Session 05: Agents
set AGENT_MODEL_PRIMARY=phi-4-mini
set AGENT_QUESTION="Why use edge AI?"

# Session 06: Pipeline
set PIPELINE_TASK="Your task here"
​

✅ 验证与测试

# Validate syntax and imports
python scripts/validate_samples.py

# Validate specific session
python scripts/validate_samples.py --session 01

# Run smoke tests
python scripts/test_samples.py --quick

# Verbose testing
python scripts/test_samples.py --verbose
​

故障排除

连接错误

# Check Foundry Local
foundry service status

# Start if needed
foundry service start
foundry model run phi-4-mini
​

导入错误

# Install missing dependencies
pip install sentence-transformers ragas datasets

# Or install all
pip install -r Workshop/requirements.txt
​

模型未找到

# List available models
foundry model ls

# Download model
foundry model download phi-4-mini
​

性能缓慢

# Use smaller model
set FOUNDRY_LOCAL_ALIAS=qwen2.5-0.5b

# Reduce benchmark rounds
set BENCH_ROUNDS=1
​

常见模式

基础聊天

from workshop_utils import chat_once

text, usage = chat_once(
    'phi-4-mini',
    messages=[{"role": "user", "content": "Hello"}],
    max_tokens=100,
    temperature=0.7
)
​

获取客户端

from workshop_utils import get_client

manager, client, model_id = get_client(
    alias='phi-4-mini',
    endpoint=None  # Auto-detect
)
​

错误处理

try:
    manager, client, model_id = get_client(alias)
except Exception as e:
    print(f"[ERROR] Failed: {e}")
    print("[INFO] Check: foundry service status")
    sys.exit(1)
​

流式传输

stream = client.chat.completions.create(
    model=model_id,
    messages=messages,
    stream=True
)

for chunk in stream:
    if chunk.choices[0].delta.content:
        print(chunk.choices[0].delta.content, end="", flush=True)
​

模型选择

资源

• SDK文档：https://github.com/microsoft/Foundry-Local/tree/main/sdk/python
• 快速参考：Workshop/FOUNDRY_SDK_QUICKREF.md

提示

1. 缓存客户端：workshop_utils会为你缓存
2. 使用较小的模型：测试时从qwen2.5-0.5b开始
3. 启用使用统计：设置SHOW_USAGE=1以跟踪令牌使用情况
4. 批量处理：顺序处理多个提示
5. 降低max_tokens：减少延迟以获得快速响应

示例工作流

测试所有内容

python scripts/validate_samples.py
python scripts/test_samples.py --quick
​

基准测试模型

cd samples
set BENCH_MODELS=phi-4-mini,qwen2.5-0.5b
set BENCH_ROUNDS=3
python -m session03.benchmark_oss_models
​

RAG管道

cd samples
set RAG_QUESTION="What is RAG?"
python -m session02.rag_pipeline
​

多代理系统

cd samples
set AGENT_QUESTION="Why edge AI for healthcare?"
python -m session05.agents_orchestrator
​

快速帮助：在samples目录中运行任何示例并使用--help，或查看文档字符串：

python -c "import session01.chat_bootstrap; help(session01.chat_bootstrap)"
​

所有示例已于2025年10月更新，采用Foundry Local SDK最佳实践 ✨

免责声明：
本文档使用AI翻译服务Co-op Translator进行翻译。尽管我们努力确保翻译的准确性，但请注意，自动翻译可能包含错误或不准确之处。原始语言的文档应被视为权威来源。对于重要信息，建议使用专业人工翻译。我们不对因使用此翻译而产生的任何误解或误读承担责任。