PyPI - langroid - Versions diffs - 0.1.207__tar.gz → 0.1.209__tar.gz - Mend

@@ -927,6 +927,13 @@ class DocChatAgent(ChatAgent):
         """
         if self.vecdb is None or self.config.n_neighbor_chunks == 0:
             return docs_scores
+        if len(docs_scores) == 0:
+            return []
+        if set(docs_scores[0][0].__fields__) != {"content", "metadata"}:
+            # Do not add context window when there are other fields besides just
+            # content and metadata, since we do not know how to set those other fields
+            # for newly created docs with combined content.
+            return docs_scores
         return self.vecdb.add_context_window(docs_scores, self.config.n_neighbor_chunks)
     def get_semantic_search_results(
@@ -1005,11 +1012,11 @@ class DocChatAgent(ChatAgent):
         # keep only docs with unique d.id()
         id2doc_score = {d.id(): (d, s) for d, s in docs_and_scores}
         docs_and_scores = list(id2doc_score.values())
-        passages = [
-            Document(content=d.content, metadata=d.metadata)
-            for (d, _) in docs_and_scores
-        ]
+        passages = [d for (d, _) in docs_and_scores]
+        # passages = [
+        #     Document(content=d.content, metadata=d.metadata)
+        #     for (d, _) in docs_and_scores
+        # ]
         if self.config.use_bm25_search:
             docs_scores = self.get_similar_chunks_bm25(query, retrieval_multiple)
@@ -1138,13 +1145,19 @@ class DocChatAgent(ChatAgent):
             input_map=lambda msg: msg.content,
             output_map=lambda ans: ans.content if ans is not None else NO_ANSWER,
         )
-        metadatas = [P.metadata for P in passages]
-        # return with metadata so we can use it downstream, e.g. to cite sources
-        return [
-            Document(content=e, metadata=m)
-            for e, m in zip(extracts, metadatas)
-            if (e != NO_ANSWER and len(e) > 0)
-        ]
+        # Caution: Retain ALL other fields in the Documents (which could be
+        # other than just `content` and `metadata`), while simply replacing
+        # `content` with the extracted portions
+        passage_extracts = []
+        for p, e in zip(passages, extracts):
+            if e == NO_ANSWER or len(e) == 0:
+                continue
+            p_copy = p.copy()
+            p_copy.content = e
+            passage_extracts.append(p_copy)
+        return passage_extracts
     def answer_from_docs(self, query: str) -> ChatDocument:
         """

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: langroid
-Version: 0.1.207
+Version: 0.1.209
 Summary: Harness LLMs with Multi-Agent Programming
 License: MIT
 Author: Prasad Chalasani

@@ -64,6 +64,14 @@ def find_fuzzy_matches_in_docs(
                 break
     if words_after is None and words_before is None:
         return orig_doc_matches
+    if len(orig_doc_matches) == 0:
+        return []
+    if set(orig_doc_matches[0].__fields__) != {"content", "metadata"}:
+        # If there are fields beyond just content and metadata,
+        # we do NOT want to create new document objects with content fields
+        # based on words_before and words_after, since we don't know how to
+        # set those other fields.
+        return orig_doc_matches
     contextual_matches = []
     for match in orig_doc_matches:

@@ -284,7 +284,8 @@ def extract_numbered_segments(s: str, specs: str) -> str:
     # Regular expression to identify numbered segments like
     # <#1#> Hello world! This is me. <#2#> How are you? <#3#> Have a good day.
-    segment_pattern = re.compile(r"<#(\d+)#> ((?:(?!<#).)+)")
+    # Note we match any character between segment markers, including newlines.
+    segment_pattern = re.compile(r"<#(\d+)#>([\s\S]*?)(?=<#\d+#>|$)")
     # Split the text into paragraphs while preserving their boundaries
     paragraphs = split_paragraphs(s)

@@ -133,10 +133,6 @@ class VectorStore(ABC):
         """Compute a result on a set of documents,
         using a dataframe calc string like `df.groupby('state')['income'].mean()`.
         """
-        # docs may be missing some fields since they may have come from
-        # DocChatAgent retrieval, so we use the ids to get the full docs
-        ids = [str(doc.id()) for doc in docs]
-        docs = self.get_documents_by_ids(ids)
         dicts = [doc.dict() for doc in docs]
         df = pd.DataFrame(dicts)

@@ -420,9 +420,8 @@ class LanceDB(VectorStore):
         docs = []
         for _id in _ids:
             results = self._lance_result_to_docs(tbl.search().where(f"id == '{_id}'"))
-            if len(results) == 0:
-                raise ValueError(f"Document with id {_id} not found")
-            docs.append(results[0])
+            if len(results) > 0:
+                docs.append(results[0])
         return docs
     def similar_texts_with_scores(

@@ -1,6 +1,6 @@
 [tool.poetry]
 name = "langroid"
-version = "0.1.207"
+version = "0.1.209"
 description = "Harness LLMs with Multi-Agent Programming"
 authors = ["Prasad Chalasani <pchalasani@gmail.com>"]
 readme = "README.md"

langroid 0.1.207__tar.gz → 0.1.209__tar.gz

langroid 0.1.207tar.gz → 0.1.209tar.gz