feat(knowledge): integrate reranking functionality into RetrieverResource

- Add reranking support to improve the quality of retrieved chunks
- Implement RerankEmbeddingFactory and RerankEmbeddingsRanker
- Update RetrieverResource to use reranking when configured
- Modify retrieve method to apply reranking if needed

packages/dbgpt-core/src/dbgpt/agent/resource/knowledge.py

@@ -5,8 +5,11 @@ from typing import TYPE_CHECKING, Any, Dict, List, Optional, Tuple, Type
 
 import cachetools
 
+from dbgpt._private.config import Config
 from dbgpt.core import Chunk
 from dbgpt.util.cache_utils import cached
+from dbgpt.rag.retriever.rerank import RerankEmbeddingsRanker
+from dbgpt.rag.embedding.embedding_factory import RerankEmbeddingFactory
 
 from .base import Resource, ResourceParameters, ResourceType
 
@@ -14,6 +17,7 @@ if TYPE_CHECKING:
     from dbgpt.rag.retriever.base import BaseRetriever
     from dbgpt.storage.vector_store.filters import MetadataFilters
 
+CFG = Config()
 
 @dataclasses.dataclass
 class RetrieverResourceParameters(ResourceParameters):
@@ -32,6 +36,12 @@ class RetrieverResource(Resource[ResourceParameters]):
         """Create a new RetrieverResource."""
         self._name = name
         self._retriever = retriever
+        app_config = CFG.SYSTEM_APP.config.configs.get("app_config")
+        rerank_embeddings = RerankEmbeddingFactory.get_instance(
+            CFG.SYSTEM_APP
+        ).create()
+        self.need_rerank = bool(app_config.models.rerankers)
+        self.reranker = RerankEmbeddingsRanker(rerank_embeddings, topk=app_config.rag.rerank_top_k)
 
     @property
     def name(self) -> str:
@@ -77,6 +87,9 @@ class RetrieverResource(Resource[ResourceParameters]):
         if not question:
             raise ValueError("Question is required for knowledge resource.")
         chunks = await self.retrieve(question)
+        if self.need_rerank and len(chunks) > 1:
+            chunks = self.reranker.rank(candidates_with_scores=chunks, query=question)
+
         content = "\n".join(
             [f"--{i}--:" + chunk.content for i, chunk in enumerate(chunks)]
         )