feat: add RAG_RERANKING_BATCH_SIZE configuration option

Add configurable reranker batch size (env var RAG_RERANKING_BATCH_SIZE,
default 32) following the same pattern as RAG_EMBEDDING_BATCH_SIZE.

- config.py: PersistentConfig for RAG_RERANKING_BATCH_SIZE
- main.py: import, state init, pass to get_reranking_function
- colbert.py: accept batch_size param in predict() (was hardcoded 32)
- utils.py: get_reranking_function passes batch_size at call time
- retrieval.py: expose in config GET/POST endpoints and ConfigForm
- Documents.svelte: add Reranking Batch Size input in admin settings

Closes #23730

backend/open_webui/retrieval/models/colbert.py

@@ -59,14 +59,14 @@ class ColBERT(BaseReranker):
 
         return normalized_scores.detach().cpu().numpy().astype(np.float32)
 
-    def predict(self, sentences):
+    def predict(self, sentences, batch_size=32):
         query = sentences[0][0]
         docs = [i[1] for i in sentences]
 
         # Embedding the documents
-        embedded_docs = self.ckpt.docFromText(docs, bsize=32)[0]
+        embedded_docs = self.ckpt.docFromText(docs, bsize=batch_size)[0]
         # Embedding the queries
-        embedded_queries = self.ckpt.queryFromText([query], bsize=32)
+        embedded_queries = self.ckpt.queryFromText([query], bsize=batch_size)
         embedded_query = embedded_queries[0]
 
         # Calculate retrieval scores for the query against all documents

backend/open_webui/retrieval/utils.py

@@ -919,7 +919,7 @@ async def generate_embeddings(
         return embeddings[0] if isinstance(text, str) else embeddings
 
 
-def get_reranking_function(reranking_engine, reranking_model, reranking_function):
+def get_reranking_function(reranking_engine, reranking_model, reranking_function, reranking_batch_size=32):
     if reranking_function is None:
         return None
     if reranking_engine == 'external':
@@ -928,7 +928,7 @@ def get_reranking_function(reranking_engine, reranking_model, reranking_function
         )
     else:
         return lambda query, documents, user=None: reranking_function.predict(
-            [(query, doc.page_content) for doc in documents]
+            [(query, doc.page_content) for doc in documents], batch_size=int(reranking_batch_size)
         )