feat(datasource):add oceanbase support (#1622)

Co-authored-by: csunny <cfqsunny@163.com>
Co-authored-by: aries_ckt <916701291@qq.com>

dbgpt/rag/embedding/embeddings.py

@@ -90,9 +90,9 @@ class HuggingFaceEmbeddings(BaseModel, Embeddings):
             ) from exc
 
         kwargs["client"] = sentence_transformers.SentenceTransformer(
-            kwargs.get("model_name"),
+            kwargs.get("model_name") or DEFAULT_MODEL_NAME,
             cache_folder=kwargs.get("cache_folder"),
-            **kwargs.get("model_kwargs"),
+            **(kwargs.get("model_kwargs") or {}),
         )
         super().__init__(**kwargs)
 

dbgpt/rag/embedding/rerank.py

@@ -3,6 +3,7 @@
 from typing import Any, Dict, List, Optional, cast
 
 import aiohttp
+import numpy as np
 import requests
 
 from dbgpt._private.pydantic import EXTRA_FORBID, BaseModel, ConfigDict, Field
@@ -16,7 +17,7 @@ class CrossEncoderRerankEmbeddings(BaseModel, RerankEmbeddings):
 
     client: Any  #: :meta private:
     model_name: str = "BAAI/bge-reranker-base"
-    max_length: Optional[int] = None
+    max_length: int
     """Max length for input sequences. Longer sequences will be truncated. If None, max
         length of the model will be used"""
     """Model name to use."""
@@ -33,9 +34,9 @@ class CrossEncoderRerankEmbeddings(BaseModel, RerankEmbeddings):
             )
 
         kwargs["client"] = CrossEncoder(
-            kwargs.get("model_name"),
+            kwargs.get("model_name", "BAAI/bge-reranker-base"),
             max_length=kwargs.get("max_length"),
-            **kwargs.get("model_kwargs"),
+            **(kwargs.get("model_kwargs") or {}),
         )
         super().__init__(**kwargs)
 
@@ -54,7 +55,9 @@ class CrossEncoderRerankEmbeddings(BaseModel, RerankEmbeddings):
         query_content_pairs = [[query, candidate] for candidate in candidates]
         _model = cast(CrossEncoder, self.client)
         rank_scores = _model.predict(sentences=query_content_pairs)
-        return rank_scores.tolist()
+        if isinstance(rank_scores, np.ndarray):
+            rank_scores = rank_scores.tolist()
+        return rank_scores
 
 
 class OpenAPIRerankEmbeddings(BaseModel, RerankEmbeddings):