langroid 0.56.8__py3-none-any.whl → 0.56.10__py3-none-any.whl

langroid/agent/special/doc_chat_agent.py

@@ -149,8 +149,8 @@ class DocChatAgentConfig(ChatAgentConfig):
     n_fuzzy_neighbor_words: int = 100  # num neighbor words to retrieve for fuzzy match
     use_fuzzy_match: bool = True
     use_bm25_search: bool = True
-    use_reciprocal_rank_fusion: bool = True  # ignored if using cross-encoder reranking
-    cross_encoder_reranking_model: str = (
+    use_reciprocal_rank_fusion: bool = False
+    cross_encoder_reranking_model: str = (  # ignored if use_reciprocal_rank_fusion=True
         "cross-encoder/ms-marco-MiniLM-L-6-v2" if has_sentence_transformers else ""
     )
     rerank_diversity: bool = True  # rerank to maximize diversity?
@@ -249,11 +249,10 @@ class DocChatAgent(ChatAgent):
         ):
             logger.warning(
                 """
-                You have set `use_reciprocal_rank_fusion` to True,
-                but it will be ignored since you have also set
-                `cross_encoder_reranking_model` to a non-empty value.
-                To use RRF (Reciprocal Rank Fusion), set
-                `cross_encoder_reranking_model` to an empty string.
+                Ignoring `cross_encoder_reranking_model` since you have set  
+                `use_reciprocal_rank_fusion` to True.
+                To use cross-encoder reranking, set
+                `use_reciprocal_rank_fusion` to False.
                 """
             )
 
@@ -315,6 +314,10 @@ class DocChatAgent(ChatAgent):
             # Note we may have used a vecdb with a config.collection_name
             # different from the agent's config.vecdb.collection_name!!
             self.vecdb.delete_collection(collection_name)
+            # Close the old vecdb before creating a new one
+            old_vecdb = self.vecdb
+            if old_vecdb and hasattr(old_vecdb, "close"):
+                old_vecdb.close()
             self.vecdb = VectorStore.create(self.vecdb.config)
         except Exception as e:
             logger.warning(
@@ -1113,7 +1116,7 @@ class DocChatAgent(ChatAgent):
                 key=lambda x: x[0],
                 reverse=True,
             )
-            passages = [d for _, d in sorted_pairs[: self.config.n_similar_chunks]]
+            passages = [d for _, d in sorted_pairs]
         return passages
 
     def rerank_with_diversity(self, passages: List[Document]) -> List[Document]:
@@ -1320,10 +1323,7 @@ class DocChatAgent(ChatAgent):
             # TODO: Add score threshold in config
             docs_scores = self.get_similar_chunks_bm25(query, retrieval_multiple)
             id2doc.update({d.id(): d for d, _ in docs_scores})
-            if (
-                self.config.cross_encoder_reranking_model == ""
-                and self.config.use_reciprocal_rank_fusion
-            ):
+            if self.config.use_reciprocal_rank_fusion:
                 # if we're not re-ranking with a cross-encoder, and have RRF enabled,
                 # instead of accumulating the bm25 results into passages,
                 # we collect these ranks for Reciprocal Rank Fusion down below.
@@ -1338,10 +1338,7 @@ class DocChatAgent(ChatAgent):
         if self.config.use_fuzzy_match:
             # TODO: Add score threshold in config
             fuzzy_match_doc_scores = self.get_fuzzy_matches(query, retrieval_multiple)
-            if (
-                self.config.cross_encoder_reranking_model == ""
-                and self.config.use_reciprocal_rank_fusion
-            ):
+            if self.config.use_reciprocal_rank_fusion:
                 # if we're not re-ranking with a cross-encoder,
                 # instead of accumulating the fuzzy match results into passages,
                 # we collect these ranks for Reciprocal Rank Fusion down below.
@@ -1357,10 +1354,8 @@ class DocChatAgent(ChatAgent):
                 # eliminate duplicate ids
                 passages = [id2doc[id] for id in id2doc.keys()]
 
-        if (
-            self.config.cross_encoder_reranking_model == ""
-            and self.config.use_reciprocal_rank_fusion
-            and (self.config.use_bm25_search or self.config.use_fuzzy_match)
+        if self.config.use_reciprocal_rank_fusion and (
+            self.config.use_bm25_search or self.config.use_fuzzy_match
         ):
             # Since we're not using cross-enocder re-ranking,
             # we need to re-order the retrieved chunks from potentially three
@@ -1382,9 +1377,9 @@ class DocChatAgent(ChatAgent):
                 # Use max_rank instead of infinity to avoid bias against
                 # single-method docs
                 max_rank = self.config.n_similar_chunks * retrieval_multiple
-                rank_semantic = id2_rank_semantic.get(id_, max_rank)
-                rank_bm25 = id2_rank_bm25.get(id_, max_rank)
-                rank_fuzzy = id2_rank_fuzzy.get(id_, max_rank)
+                rank_semantic = id2_rank_semantic.get(id_, max_rank + 1)
+                rank_bm25 = id2_rank_bm25.get(id_, max_rank + 1)
+                rank_fuzzy = id2_rank_fuzzy.get(id_, max_rank + 1)
                 c = self.config.reciprocal_rank_fusion_constant
                 reciprocal_fusion_score = (
                     1 / (rank_semantic + c) + 1 / (rank_bm25 + c) + 1 / (rank_fuzzy + c)
@@ -1421,10 +1416,14 @@ class DocChatAgent(ChatAgent):
             passages_scores = self.add_context_window(passages_scores)
             passages = [p for p, _ in passages_scores]
         # now passages can potentially have a lot of doc chunks,
-        # so we re-rank them using a cross-encoder scoring model,
+        # so we re-rank them using a cross-encoder scoring model
+        # (provided that `reciprocal_rank_fusion` is not enabled),
         # and pick top k where k = config..n_similar_chunks
         # https://www.sbert.net/examples/applications/retrieve_rerank
-        if self.config.cross_encoder_reranking_model != "":
+        if (
+            self.config.cross_encoder_reranking_model != ""
+            and not self.config.use_reciprocal_rank_fusion
+        ):
             passages = self.rerank_with_cross_encoder(query, passages)
 
         if self.config.rerank_diversity:

langroid/vector_store/qdrantdb.py

@@ -4,7 +4,7 @@ import logging
 import os
 import time
 import uuid
-from typing import TYPE_CHECKING, Dict, List, Optional, Sequence, Tuple, TypeVar
+from typing import TYPE_CHECKING, Any, Dict, List, Optional, Sequence, Tuple, TypeVar
 
 from dotenv import load_dotenv
 
@@ -143,6 +143,24 @@ class QdrantDB(VectorStore):
                 config.collection_name, replace=config.replace_collection
             )
 
+    def close(self) -> None:
+        """
+        Close the QdrantDB client and release any resources (e.g., file locks).
+        This is especially important for local storage to release the .lock file.
+        """
+        if hasattr(self.client, "close"):
+            # QdrantLocal has a close method that releases the lock
+            self.client.close()
+            logger.info(f"Closed QdrantDB connection for {self.config.storage_path}")
+
+    def __enter__(self) -> "QdrantDB":
+        """Context manager entry."""
+        return self
+
+    def __exit__(self, exc_type: Any, exc_val: Any, exc_tb: Any) -> None:
+        """Context manager exit - ensure cleanup even if an exception occurred."""
+        self.close()
+
     def clear_empty_collections(self) -> int:
         coll_names = self.list_collections()
         n_deletes = 0

{langroid-0.56.8.dist-info → langroid-0.56.10.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: langroid
-Version: 0.56.8
+Version: 0.56.10
 Summary: Harness LLMs with Multi-Agent Programming
 Author-email: Prasad Chalasani <pchalasani@gmail.com>
 License: MIT

{langroid-0.56.8.dist-info → langroid-0.56.10.dist-info}/RECORD

@@ -15,7 +15,7 @@ langroid/agent/xml_tool_message.py,sha256=oeBKnJNoGaKdtz39XoWGMTNlVyXew2MWH5lgtY
 langroid/agent/callbacks/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 langroid/agent/callbacks/chainlit.py,sha256=4rJw07NIIVTIVvksVY08h5PdLE_kRoJItjbQM0UjRn0,20962
 langroid/agent/special/__init__.py,sha256=gik_Xtm_zV7U9s30Mn8UX3Gyuy4jTjQe9zjiE3HWmEo,1273
-langroid/agent/special/doc_chat_agent.py,sha256=9Evo3d8iniCmU6Liw0lq2HElPcavHA_M9I7_kUMhrRk,68860
+langroid/agent/special/doc_chat_agent.py,sha256=PysF6K_rj_xe1yHVZyaldQkGacdHv905zmNBqOdrzYU,68815
 langroid/agent/special/doc_chat_task.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 langroid/agent/special/lance_doc_chat_agent.py,sha256=6pIqi2DF-MvYYN3-blsdUgulYnOBTl7I21T7wPAt1zM,10413
 langroid/agent/special/lance_tools.py,sha256=qS8x4wi8mrqfbYV2ztFzrcxyhHQ0ZWOc-zkYiH7awj0,2105
@@ -135,9 +135,9 @@ langroid/vector_store/lancedb.py,sha256=Qd20gKjWozPWfW5-D66J6U8dSrJo1yl-maj6s1lb
 langroid/vector_store/meilisearch.py,sha256=6frB7GFWeWmeKzRfLZIvzRjllniZ1cYj3HmhHQICXLs,11663
 langroid/vector_store/pineconedb.py,sha256=otxXZNaBKb9f_H75HTaU3lMHiaR2NUp5MqwLZXpEY9M,14994
 langroid/vector_store/postgres.py,sha256=wHPtIi2qM4fhO4pMQr95pz1ZCe7dTb2hxl4VYspGZoA,16104
-langroid/vector_store/qdrantdb.py,sha256=O6dSBoDZ0jzfeVBd7LLvsXu083xs2fxXtPa9gGX3JX4,18443
+langroid/vector_store/qdrantdb.py,sha256=ZYrT9mxoUCx_67Qzb5xnkWuFG12rfe30yAg4NgG2ueA,19168
 langroid/vector_store/weaviatedb.py,sha256=Yn8pg139gOy3zkaPfoTbMXEEBCiLiYa1MU5d_3UA1K4,11847
-langroid-0.56.8.dist-info/METADATA,sha256=waeMwWJeG_cwWJo8btFzQDdj1EbhDN2Ww1_YfXvRCBM,65744
-langroid-0.56.8.dist-info/WHEEL,sha256=qtCwoSJWgHk21S1Kb4ihdzI2rlJ1ZKaIurTj_ngOhyQ,87
-langroid-0.56.8.dist-info/licenses/LICENSE,sha256=EgVbvA6VSYgUlvC3RvPKehSg7MFaxWDsFuzLOsPPfJg,1065
-langroid-0.56.8.dist-info/RECORD,,
+langroid-0.56.10.dist-info/METADATA,sha256=QYPsEwh24uWqM4OwExH1tSmsWlnj-cyFRkTAXd2Rl64,65745
+langroid-0.56.10.dist-info/WHEEL,sha256=qtCwoSJWgHk21S1Kb4ihdzI2rlJ1ZKaIurTj_ngOhyQ,87
+langroid-0.56.10.dist-info/licenses/LICENSE,sha256=EgVbvA6VSYgUlvC3RvPKehSg7MFaxWDsFuzLOsPPfJg,1065
+langroid-0.56.10.dist-info/RECORD,,