langroid 0.56.7__py3-none-any.whl → 0.56.9__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.
                 """
             )
 
@@ -1113,7 +1112,7 @@ 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 +1319,7 @@ 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 +1334,7 @@ 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 +1350,8 @@ 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 +1373,9 @@ 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 +1412,14 @@ 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/agent/tools/task_tool.py

@@ -24,6 +24,7 @@ class TaskTool(ToolMessage):
     will be generated.
     """
 
+    # TODO: setting up termination conditions of sub-task needs to be improved
     request: str = "task_tool"
     purpose: str = """
         <HowToUse>
@@ -77,6 +78,7 @@ class TaskTool(ToolMessage):
         To disable all tools, set it to a singleton list containing 'NONE'
         """,
     )
+    # TODO: ensure valid model name
     model: str = Field(
         default=None,
         description="""
@@ -183,5 +185,7 @@ class TaskTool(ToolMessage):
         """
         task = self._set_up_task(agent)
         # Run the task on the prompt, and return the result
+        # TODO eventually allow the various task setup configs,
+        #  including termination conditions
         result = await task.run_async(self.prompt, turns=self.max_iterations or 10)
         return result

langroid/language_models/base.py

@@ -78,7 +78,7 @@ class LLMConfig(BaseSettings):
     completion_model: str = ""
     temperature: float = 0.0
     chat_context_length: int | None = None
-    async_stream_quiet: bool = True  # suppress streaming output in async mode?
+    async_stream_quiet: bool = False  # suppress streaming output in async mode?
     completion_context_length: int | None = None
     # if input length + max_output_tokens > context length of model,
     # we will try shortening requested output

{langroid-0.56.7.dist-info → langroid-0.56.9.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: langroid
-Version: 0.56.7
+Version: 0.56.9
 Summary: Harness LLMs with Multi-Agent Programming
 Author-email: Prasad Chalasani <pchalasani@gmail.com>
 License: MIT

{langroid-0.56.7.dist-info → langroid-0.56.9.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=s3cQTUzbfuLiDOG8x5aRyZd1Ql-b1alHWOiQ4b5wgF8,68628
 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
@@ -54,7 +54,7 @@ langroid/agent/tools/recipient_tool.py,sha256=dr0yTxgNEIoxUYxH6TtaExC4G_8WdJ0xGo
 langroid/agent/tools/retrieval_tool.py,sha256=zcAV20PP_6VzSd-UE-IJcabaBseFL_QNz59Bnig8-lE,946
 langroid/agent/tools/rewind_tool.py,sha256=XAXL3BpNhCmBGYq_qi_sZfHJuIw7NY2jp4wnojJ7WRs,5606
 langroid/agent/tools/segment_extract_tool.py,sha256=__srZ_VGYLVOdPrITUM8S0HpmX4q7r5FHWMDdHdEv8w,1440
-langroid/agent/tools/task_tool.py,sha256=VOHWv8uFPRczf6qp8YZPpUirkCqE6YVhN6jgXCvdpe0,7102
+langroid/agent/tools/task_tool.py,sha256=2qg3oRDYjF93pi7J0gwgBjPL4R9ByKH6yMwD-Cx8voo,7325
 langroid/agent/tools/tavily_search_tool.py,sha256=soI-j0HdgVQLf09wRQScaEK4b5RpAX9C4cwOivRFWWI,1903
 langroid/agent/tools/mcp/__init__.py,sha256=DJNM0VeFnFS3pJKCyFGggT8JVjVu0rBzrGzasT1HaSM,387
 langroid/agent/tools/mcp/decorators.py,sha256=h7dterhsmvWJ8q4mp_OopmuG2DF71ty8cZwOyzdDZuk,1127
@@ -73,7 +73,7 @@ langroid/embedding_models/protoc/embeddings_pb2.pyi,sha256=UkNy7BrNsmQm0vLb3NtGX
 langroid/embedding_models/protoc/embeddings_pb2_grpc.py,sha256=9dYQqkW3JPyBpSEjeGXTNpSqAkC-6FPtBHyteVob2Y8,2452
 langroid/language_models/__init__.py,sha256=3aD2qC1lz8v12HX4B-dilv27gNxYdGdeu1QvDlkqqHs,1095
 langroid/language_models/azure_openai.py,sha256=SW0Fp_y6HpERr9l6TtF6CYsKgKwjUf_hSL_2mhTV4wI,5034
-langroid/language_models/base.py,sha256=253xcwXZ0yxSQ1W4SR50tAPZKCDc35yyU1o35EqB9b8,28484
+langroid/language_models/base.py,sha256=OlPgmhQS2o3Y5DLoO1IEBUp0kIOeQdYsZsd25sz7DY8,28485
 langroid/language_models/config.py,sha256=9Q8wk5a7RQr8LGMT_0WkpjY8S4ywK06SalVRjXlfCiI,378
 langroid/language_models/mock_lm.py,sha256=tA9JpURznsMZ59iRhFYMmaYQzAc0D0BT-PiJIV58sAk,4079
 langroid/language_models/model_info.py,sha256=0e011vJZMi7XU9OkKT6doxlybrNJfMlP54klLDDNgFg,14939
@@ -137,7 +137,7 @@ langroid/vector_store/pineconedb.py,sha256=otxXZNaBKb9f_H75HTaU3lMHiaR2NUp5MqwLZ
 langroid/vector_store/postgres.py,sha256=wHPtIi2qM4fhO4pMQr95pz1ZCe7dTb2hxl4VYspGZoA,16104
 langroid/vector_store/qdrantdb.py,sha256=O6dSBoDZ0jzfeVBd7LLvsXu083xs2fxXtPa9gGX3JX4,18443
 langroid/vector_store/weaviatedb.py,sha256=Yn8pg139gOy3zkaPfoTbMXEEBCiLiYa1MU5d_3UA1K4,11847
-langroid-0.56.7.dist-info/METADATA,sha256=czhMl375GcCjQoavgVE7vQUbMNkan6D4_S83glup2to,65744
-langroid-0.56.7.dist-info/WHEEL,sha256=qtCwoSJWgHk21S1Kb4ihdzI2rlJ1ZKaIurTj_ngOhyQ,87
-langroid-0.56.7.dist-info/licenses/LICENSE,sha256=EgVbvA6VSYgUlvC3RvPKehSg7MFaxWDsFuzLOsPPfJg,1065
-langroid-0.56.7.dist-info/RECORD,,
+langroid-0.56.9.dist-info/METADATA,sha256=uppdkleoPy1RXOJ6sGIRfFDrYhcSUKAZ_1rfjaYErA8,65744
+langroid-0.56.9.dist-info/WHEEL,sha256=qtCwoSJWgHk21S1Kb4ihdzI2rlJ1ZKaIurTj_ngOhyQ,87
+langroid-0.56.9.dist-info/licenses/LICENSE,sha256=EgVbvA6VSYgUlvC3RvPKehSg7MFaxWDsFuzLOsPPfJg,1065
+langroid-0.56.9.dist-info/RECORD,,