RAG: Get relevance score when using Reranker Module

[ChenJiahaoST]

Hi, I just found that I can't get relevance scores when using reranker(ModuleReranker), I can just get a sorted_indices like "[0, 3, 1, 2]" and get topk DocNodes

reranker_model = Reranker('ModuleReranker', model=bge-reranker-large
reranker_model.start()
nodes = [
        DocNode(text="深度学习是..."),
        DocNode(text="机器学习是..."),
        DocNode(text="自然语言处理是..."),
        DocNode(text="计算机视觉是..."),
    ]
query = "深度学习是什么？"
reranker_model(nodes, query)

But when I use other reranker model , I can get a series of relevance socre telling the relevance between node and query so that I can set a threshold to filter some bad nodes to optimize the generation performance of my RAG system.

like (https://jina.ai/reranker/):

{
  "model": "jina-reranker-v2-base-multilingual",
  "usage": {
    "total_tokens": 48
  },
  "results": [
    {
      "index": 0,
      "document": {
        "text": "深度学习是..."
      },
      "relevance_score": 0.6261242032051086
    },
    {
      "index": 2,
      "document": {
        "text": "自然语言处理是..."
      },
      "relevance_score": 0.1037486344575882
    },
    {
      "index": 1,
      "document": {
        "text": "机器学习是..."
      },
      "relevance_score": 0.09009299427270889
    },
    {
      "index": 3,
      "document": {
        "text": "计算机视觉是..."
      },
      "relevance_score": 0.08035746961832047
    }
  ]
}

Can I get the relevance score like this by using LazyLLM's Rerankers? Not just topk

[wzh1994]

这个功能在计划中，issue见#155 的第8条，我们会尽快支持