RankGPT¶

The ‘RankGPT’ module is a reranker using RankGPT. RankGPT uses GPT or other LLM model to reranking passages based on their relevance to a given query.

Module Parameters¶

llm: Default LLM model is gpt-3.5-turbo-16k, but you can use another LLM model to use RankGPT. The usage on the config YAML file for setting custom LLM is the same as llama_index_llm module. More details can be found on the Local Model page. Set llm as the name of LlamaIndex class name, and write the rest of parameters.
verbose: The verbosity of the RankGPT module. Default is False.
batch: Batch size. Since this module is using LLM model, choose wisely to prevent OOM or token limit error. Default is 16.
rankgpt_rerank_prompt: The rerank prompt. Default is RankGPT default prompt.

Example config.yaml¶

modules:
  - module_type: rankgpt
    llm: openai
    model: gpt-4
    temperature: 0.5
    verbose: False
    batch: 8