Best Practices

Prompt Engineering

Use System Messages

Set the behavior and tone with system messages:

messages = [
    {"role": "system", "content": "You are an expert Python developer. Answer with code examples."},
    {"role": "user", "content": "How do I sort a dictionary by value?"},
]

Be Specific

Chinese models respond well to detailed, structured prompts:

❌ Bad:  "Write code"
✅ Good: "Write a Python function that takes a list of integers and returns the top 3 most frequent values"

Performance Optimization

Set max_tokens

Limit response length to reduce cost and latency:

response = client.chat.completions.create(
    model="deepseek-chat",
    messages=[{"role": "user", "content": "Summarize: ..."}],
    max_tokens=200,  # Short summary
)

Use Streaming for UX

For chat interfaces, always stream to show tokens as they arrive:

stream = client.chat.completions.create(
    model="deepseek-chat",
    messages=messages,
    stream=True,
)

Cost Management

Track Token Usage

Always log usage to monitor costs:

response = client.chat.completions.create(...)
print(f"Cost: {response.usage.total_tokens} tokens")
# prompt_tokens, completion_tokens, total_tokens

Choose the Right Model

Task	Best Model	Why
Chat / Support	deepseek-chat	Best price-performance
Math / Logic	deepseek-reasoner	Specialized reasoning
Translation	qwen-max	Multilingual optimized
Chinese content	glm-4	Chinese-native model
Budget tasks	qwen-plus	Cheap, still capable

Error Handling

Always handle API errors gracefully:

from openai import (
    APIError, RateLimitError, APIConnectionError, AuthenticationError
)

try:
    response = client.chat.completions.create(...)
except AuthenticationError:
    print("Invalid API key")
except RateLimitError:
    print("Too many requests — backing off")
    time.sleep(10)
except APIConnectionError:
    print("Network issue — retrying")
except APIError as e:
    print(f"API error: {e}")

Security

• Use environment variables, never hardcode keys
• Rotate keys periodically
• Use separate keys for development and production
• Never expose keys in client-side code or public repos