> ## Documentation Index > Fetch the complete documentation index at: https://docs.tokenlab.sh/llms.txt > Use this file to discover all available pages before exploring further. # 最佳實務 > 最佳化您的 TokenLab API 使用方式，以提升成本效益、效能與可靠性 ## 模型選擇 選擇合適的模型會大幅影響成本與品質。 ### 依任務類型的建議 | 任務 | 建議模型 | 原因 | | --------- | ------------------------------------------------- | ------------- | | **簡單問答** | `gpt-5-mini`, `gemini-3.5-flash` | 快速、便宜、已足夠應付需求 | | **複雜推理** | `gpt-5.4`, `claude-opus-4-6`, `deepseek-r1` | 邏輯與規劃能力更佳 | | **程式撰寫** | `claude-sonnet-4-6`, `gpt-4o`, `deepseek-v3-2` | 針對程式碼進行最佳化 | | **創意寫作** | `claude-sonnet-4-6`, `gpt-4o` | 文字表達品質更佳 | | **視覺/影像** | `gpt-4o`, `claude-sonnet-4-6`, `gemini-3.5-flash` | 原生支援視覺能力 | | **長上下文** | `gemini-2.5-pro`, `claude-sonnet-4-6` | 1M+ token 視窗 | | **成本敏感** | `gpt-5-mini`, `gemini-3.5-flash`, `deepseek-v3-2` | 最佳性價比 | ### 成本層級 ``` $$$$ Premium: gpt-5.4, claude-opus-4-6 $$$ Standard: claude-sonnet-4-6, gpt-4o $$ Budget: gpt-5-mini, gemini-3.5-flash $ Economy: deepseek-v3-2, deepseek-r1 ``` ## 成本最佳化 ### 1. 優先使用較小型的模型 ```python theme={null} def smart_query(question: str, complexity: str = "auto"): """Use cheaper models for simple tasks.""" if complexity == "simple": model = "gpt-5-mini" elif complexity == "complex": model = "gpt-4o" else: # Start cheap, escalate if needed model = "gpt-5-mini" response = client.chat.completions.create( model=model, messages=[{"role": "user", "content": question}] ) return response ``` ### 2. 設定 max\_tokens 請務必設定合理的 `max_tokens` 上限： ```python theme={null} # ❌ Bad: No limit, could generate thousands of tokens response = client.chat.completions.create( model="gpt-4o", messages=[{"role": "user", "content": "Summarize this article"}] ) # ✅ Good: Limit response length response = client.chat.completions.create( model="gpt-4o", messages=[{"role": "user", "content": "Summarize this article"}], max_tokens=500 # Reasonable limit for a summary ) ``` ### 3. 最佳化 Prompt ```python theme={null} # ❌ Verbose prompt (more input tokens) prompt = """ I would like you to please help me by analyzing the following text and providing a comprehensive summary of the main points. Please be thorough but also concise in your response. The text is as follows: {text} """ # ✅ Concise prompt (fewer tokens) prompt = "Summarize the key points:\n{text}" ``` ### 4. 批次處理相似請求 ```python theme={null} # ❌ Many small requests for question in questions: response = client.chat.completions.create( model="gpt-4o", messages=[{"role": "user", "content": question}] ) # ✅ Fewer larger requests combined_prompt = "\n".join([f"{i+1}. {q}" for i, q in enumerate(questions)]) response = client.chat.completions.create( model="gpt-4o", messages=[{"role": "user", "content": f"Answer each question:\n{combined_prompt}"}] ) ``` ## 效能最佳化 ### 5. 為 UX 使用串流回應 串流回應可改善使用者感受到的效能： ```python theme={null} stream = client.chat.completions.create( model="gpt-4o", messages=[{"role": "user", "content": "Write a long essay"}], stream=True ) for chunk in stream: if chunk.choices[0].delta.content: print(chunk.choices[0].delta.content, end="", flush=True) ``` ### 6. 互動式使用情境選擇快速模型 | 使用情境 | 建議 | 延遲 | | ------ | -------------------------------- | ---------------- | | 聊天 UI | `gpt-5-mini`, `gemini-3.5-flash` | \~200ms 首個 token | | Tab 補全 | `claude-haiku-4-5` | \~150ms 首個 token | | 背景處理 | `gpt-4o`, `claude-sonnet-4-6` | \~500ms 首個 token | ### 7. 設定逾時 ```python theme={null} client = OpenAI( api_key="sk-your-key", base_url="https://api.tokenlab.sh/v1", timeout=60.0 # 60 second timeout ) ``` ## 可靠性 ### 8. 實作重試機制 ```python theme={null} import time from openai import RateLimitError, APIError def chat_with_retry(messages, max_retries=3): for attempt in range(max_retries): try: return client.chat.completions.create( model="gpt-4o", messages=messages ) except RateLimitError: wait = 2 ** attempt print(f"Rate limited, waiting {wait}s...") time.sleep(wait) except APIError as e: if attempt == max_retries - 1: raise time.sleep(1) raise Exception("Max retries exceeded") ``` ### 9. 妥善處理錯誤 ```python theme={null} from openai import APIError, AuthenticationError, RateLimitError try: response = client.chat.completions.create(...) except AuthenticationError: # Check API key notify_admin("Invalid API key") except RateLimitError: # Queue for later or use backup add_to_queue(request) except APIError as e: if e.status_code == 402: notify_admin("Balance low") elif e.status_code >= 500: # Server error, retry later schedule_retry(request) ``` ### 10. 使用備援模型 ```python theme={null} FALLBACK_CHAIN = ["gpt-4o", "claude-sonnet-4-6", "gemini-3.5-flash"] def chat_with_fallback(messages): for model in FALLBACK_CHAIN: try: return client.chat.completions.create( model=model, messages=messages ) except APIError: continue raise Exception("All models failed") ``` ## 安全性 ### 11. 保護 API Keys ```python theme={null} # ❌ Never hardcode keys client = OpenAI(api_key="sk-abc123...") # ✅ Use environment variables import os client = OpenAI(api_key=os.environ["TOKENLAB_API_KEY"]) ``` ### 12. 驗證使用者輸入 ```python theme={null} def validate_message(content: str) -> bool: """Validate user input before sending to API.""" if len(content) > 100000: raise ValueError("Message too long") # Add other validation as needed return True ``` ### 13. 設定 API Key 限制 為以下用途建立具備支出上限的獨立 API keys： * 開發/測試 * 正式環境 * 不同應用程式 ## 監控 ### 14. 追蹤使用量 請定期檢查您的 dashboard，以掌握： * 各模型的 token 使用量 * 成本明細 * 快取命中率 * 錯誤率 ### 15. 記錄重要指標 ```python theme={null} import logging response = client.chat.completions.create(...) logging.info({ "model": response.model, "prompt_tokens": response.usage.prompt_tokens, "completion_tokens": response.usage.completion_tokens, "total_tokens": response.usage.total_tokens, }) ``` ### 16. 設定警示 在您的 dashboard 中設定低餘額警示，以避免服務中斷。 ## 檢查清單 * [ ] 為每項任務使用適當的模型 * [ ] 設定 max\_tokens 上限 * [ ] Prompt 保持精簡 * [ ] 在適當情況下啟用快取 * [ ] 批次處理相似請求 * [ ] 為互動式 UX 使用串流回應 * [ ] 即時使用情境採用快速模型 * [ ] 已設定逾時 * [ ] 已實作重試邏輯 * [ ] 已建立錯誤處理機制 * [ ] 已設定備援模型 * [ ] API keys 存放於環境變數中 * [ ] 輸入驗證 * [ ] 為 dev/prod 使用獨立 keys * [ ] 已設定支出上限