add different chunk tables and emb models

app.py

@@ -44,7 +44,9 @@ def bot(history, api_kind, chunk_table, embedding_model, llm_model, eross_encode
     # Retrieve documents relevant to query
     document_start = perf_counter()
 
-    documents = retrieve(query, TOP_K)
+    #documents = retrieve(query, TOP_K)
+    documents = retrieve(query, top_k_param, chunk_table, embedding_model)
+
 
     document_time = perf_counter() - document_start
     logger.info(f'Finished Retrieving documents in {round(document_time, 2)} seconds...')
@@ -118,10 +120,11 @@ with gr.Blocks() as demo:
             )    
     eross_encoder = gr.Radio(
                 choices=[
+                    "None"
                     "BAAI/bge-reranker-large",
                     "cross-encoder/ms-marco-MiniLM-L-6-v2",
                 ],
-                value="cross-encoder/ms-marco-MiniLM-L-6-v2",
+                value="None",
                 label='Cross-encoder model'
             )
     top_k_param = gr.Radio(

backend/semantic_search.py

@@ -6,15 +6,17 @@ from sentence_transformers import SentenceTransformer
 
 db = lancedb.connect(".lancedb")
 
-TABLE = db.open_table(os.getenv("TABLE_NAME"))
+#TABLE = db.open_table(os.getenv("TABLE_NAME"))
 VECTOR_COLUMN = os.getenv("VECTOR_COLUMN", "vector")
 TEXT_COLUMN = os.getenv("TEXT_COLUMN", "text")
 BATCH_SIZE = int(os.getenv("BATCH_SIZE", 32))
 
-retriever = SentenceTransformer(os.getenv("EMB_MODEL"))
+#retriever = SentenceTransformer(os.getenv("EMB_MODEL"))
 
 
-def retrieve(query, k):
+def retrieve(query, k, table_name, emb_name):
+    TABLE = db.open_table(table_name)
+    retriever = SentenceTransformer(emb_name)
     query_vec = retriever.encode(query)
     try:
         documents = TABLE.search(query_vec, vector_column_name=VECTOR_COLUMN).limit(k).to_list()