使用LiteLLM实现模型回退

以下是如何使用LiteLLM在三个LLM提供商（OpenAI、Anthropic、Azure）之间实现模型回退的方法。

1. 安装LiteLLM

!pip install litellm

2. 基本回退代码

import litellm
from litellm import embedding, completion

# 设置环境变量
os.environ["OPENAI_API_KEY"] = ""
os.environ["ANTHROPIC_API_KEY"] = ""
os.environ["AZURE_API_KEY"] = ""
os.environ["AZURE_API_BASE"] = ""
os.environ["AZURE_API_VERSION"] = ""

model_fallback_list = ["claude-instant-1", "gpt-3.5-turbo", "chatgpt-test"]

user_message = "Hello, how are you?"
messages = [{ "content": user_message,"role": "user"}]

for model in model_fallback_list:
  try:
      response = completion(model=model, messages=messages)
  except Exception as e:
      print(f"error occurred: {traceback.format_exc()}")

3. 上下文窗口异常

LiteLLM为上下文窗口超出错误提供了InvalidRequestError类的子类（文档）。

基于上下文窗口异常实现模型回退。

LiteLLM还公开了一个get_max_tokens()函数，您可以使用它来识别已超出的上下文窗口限制。

import litellm
from litellm import completion, ContextWindowExceededError, get_max_tokens

# 设置环境变量
os.environ["OPENAI_API_KEY"] = ""
os.environ["COHERE_API_KEY"] = ""
os.environ["ANTHROPIC_API_KEY"] = ""
os.environ["AZURE_API_KEY"] = ""
os.environ["AZURE_API_BASE"] = ""
os.environ["AZURE_API_VERSION"] = ""

context_window_fallback_list = [{"model":"gpt-3.5-turbo-16k", "max_tokens": 16385}, {"model":"gpt-4-32k", "max_tokens": 32768}, {"model": "claude-instant-1", "max_tokens":100000}]

user_message = "Hello, how are you?"
messages = [{ "content": user_message,"role": "user"}]

initial_model = "command-nightly"
try:
    response = completion(model=initial_model, messages=messages)
except ContextWindowExceededError as e:
    model_max_tokens = get_max_tokens(model)
    for model in context_window_fallback_list:
        if model_max_tokens < model["max_tokens"]
        try:
            response = completion(model=model["model"], messages=messages)
            return response
        except ContextWindowExceededError as e:
            model_max_tokens = get_max_tokens(model["model"])
            continue

print(response)