Hello Hacker News,

Stop relying on benchmarks and easily test LLMs in production. Try it here: https://admin.litellm.ai/

LiteLLM allows you to simplify calling any LLM as a drop in replacement for gpt-3.5-turbo

We're launching `completion_with_split_tests` to easily A/B test all LLMs.

Example usage - 1 function: completion_with_split_tests(

  models={
 "claude-2": 0.4, 
 "gpt-3.5-turbo": 0.6
  }, 

  messages=messages,

  temperature=temperature

)

For each completion call we allow you to:

- Control/Modify LLM configs (prompt, temperature, max_tokens etc without needing to edit code)

- Easily swap in/out 100+ LLMs without redeploying code

- View Input/Outputs for each LLM on our UI

- Retry requests with an alternate LLM

Happy completion()!