langroid 0.12.0__py3-none-any.whl → 0.13.0__py3-none-any.whl

langroid/agent/special/doc_chat_agent.py

@@ -49,7 +49,6 @@ from langroid.parsing.search import (
 from langroid.parsing.table_loader import describe_dataframe
 from langroid.parsing.url_loader import URLLoader
 from langroid.parsing.urls import get_list_from_user, get_urls_paths_bytes_indices
-from langroid.parsing.utils import batched
 from langroid.prompts.prompts_config import PromptsConfig
 from langroid.prompts.templates import SUMMARY_ANSWER_PROMPT_GPT4
 from langroid.utils.constants import NO_ANSWER
@@ -137,7 +136,6 @@ class DocChatAgentConfig(ChatAgentConfig):
     rerank_diversity: bool = True  # rerank to maximize diversity?
     rerank_periphery: bool = True  # rerank to avoid Lost In the Middle effect?
     rerank_after_adding_context: bool = True  # rerank after adding context window?
-    embed_batch_size: int = 500  # get embedding of at most this many at a time
     cache: bool = True  # cache results
     debug: bool = False
     stream: bool = True  # allow streaming where needed
@@ -400,7 +398,11 @@ class DocChatAgent(ChatAgent):
         if split:
             docs = self.parser.split(docs)
         else:
-            self.parser.add_window_ids(docs)
+            if self.config.n_neighbor_chunks > 0:
+                self.parser.add_window_ids(docs)
+            # we're not splitting, so we mark each doc as a chunk
+            for d in docs:
+                d.metadata.is_chunk = True
         if self.vecdb is None:
             raise ValueError("VecDB not set")
 
@@ -422,10 +424,9 @@ class DocChatAgent(ChatAgent):
                         + d.content
                     )
         docs = docs[: self.config.parsing.max_chunks]
-        # add embeddings in batches, to stay under limit of embeddings API
-        batches = list(batched(docs, self.config.embed_batch_size))
-        for batch in batches:
-            self.vecdb.add_documents(batch)
+        # vecdb should take care of adding docs in batches;
+        # batching can be controlled via vecdb.config.batch_size
+        self.vecdb.add_documents(docs)
         self.original_docs_length = self.doc_length(docs)
         self.setup_documents(docs, filter=self.config.filter)
         return len(docs)
@@ -894,7 +895,9 @@ class DocChatAgent(ChatAgent):
             )
         return docs_scores
 
-    def get_fuzzy_matches(self, query: str, multiple: int) -> List[Document]:
+    def get_fuzzy_matches(
+        self, query: str, multiple: int
+    ) -> List[Tuple[Document, float]]:
         # find similar docs using fuzzy matching:
         # these may sometimes be more likely to contain a relevant verbatim extract
         with status("[cyan]Finding fuzzy matches in chunks..."):
@@ -909,8 +912,8 @@ class DocChatAgent(ChatAgent):
                 self.chunked_docs,
                 self.chunked_docs_clean,
                 k=self.config.parsing.n_similar_docs * multiple,
-                words_before=self.config.n_fuzzy_neighbor_words,
-                words_after=self.config.n_fuzzy_neighbor_words,
+                words_before=self.config.n_fuzzy_neighbor_words or None,
+                words_after=self.config.n_fuzzy_neighbor_words or None,
             )
         return fuzzy_match_docs
 
@@ -1127,12 +1130,14 @@ class DocChatAgent(ChatAgent):
         # ]
 
         if self.config.use_bm25_search:
+            # TODO: Add score threshold in config
             docs_scores = self.get_similar_chunks_bm25(query, retrieval_multiple)
             passages += [d for (d, _) in docs_scores]
 
         if self.config.use_fuzzy_match:
-            fuzzy_match_docs = self.get_fuzzy_matches(query, retrieval_multiple)
-            passages += fuzzy_match_docs
+            # TODO: Add score threshold in config
+            fuzzy_match_doc_scores = self.get_fuzzy_matches(query, retrieval_multiple)
+            passages += [d for (d, _) in fuzzy_match_doc_scores]
 
         # keep unique passages
         id2passage = {p.id(): p for p in passages}

langroid/parsing/search.py

@@ -27,7 +27,7 @@ def find_fuzzy_matches_in_docs(
     k: int,
     words_before: int | None = None,
     words_after: int | None = None,
-) -> List[Document]:
+) -> List[Tuple[Document, float]]:
     """
     Find approximate matches of the query in the docs and return surrounding
     characters.
@@ -35,6 +35,7 @@ def find_fuzzy_matches_in_docs(
     Args:
         query (str): The search string.
         docs (List[Document]): List of Document objects to search through.
+        docs_clean (List[Document]): List of Document objects with cleaned content.
         k (int): Number of best matches to return.
         words_before (int|None): Number of words to include before each match.
             Default None => return max
@@ -42,8 +43,7 @@ def find_fuzzy_matches_in_docs(
             Default None => return max
 
     Returns:
-        List[Document]: List of Documents containing the matches,
-            including the given number of words around the match.
+        List[Tuple[Document,float]]: List of (Document, score) tuples.
     """
     if len(docs) == 0:
         return []
@@ -54,19 +54,19 @@ def find_fuzzy_matches_in_docs(
         scorer=fuzz.partial_ratio,
     )
 
-    real_matches = [m for m, score in best_matches if score > 50]
+    real_matches = [(m, score) for m, score in best_matches if score > 50]
     # find the original docs that corresponding to the matches
     orig_doc_matches = []
-    for i, m in enumerate(real_matches):
+    for i, (m, s) in enumerate(real_matches):
         for j, doc_clean in enumerate(docs_clean):
             if m in doc_clean.content:
-                orig_doc_matches.append(docs[j])
+                orig_doc_matches.append((docs[j], s))
                 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 set(orig_doc_matches[0][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
@@ -74,7 +74,7 @@ def find_fuzzy_matches_in_docs(
         return orig_doc_matches
 
     contextual_matches = []
-    for match in orig_doc_matches:
+    for match, score in orig_doc_matches:
         choice_text = match.content
         contexts = []
         while choice_text != "":
@@ -89,9 +89,12 @@ def find_fuzzy_matches_in_docs(
             choice_text = " ".join(words[end_pos:])
         if len(contexts) > 0:
             contextual_matches.append(
-                Document(
-                    content=" ... ".join(contexts),
-                    metadata=match.metadata,
+                (
+                    Document(
+                        content=" ... ".join(contexts),
+                        metadata=match.metadata,
+                    ),
+                    score,
                 )
             )
 

{langroid-0.12.0.dist-info → langroid-0.13.0.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: langroid
-Version: 0.12.0
+Version: 0.13.0
 Summary: Harness LLMs with Multi-Agent Programming
 License: MIT
 Author: Prasad Chalasani

{langroid-0.12.0.dist-info → langroid-0.13.0.dist-info}/RECORD

@@ -10,7 +10,7 @@ langroid/agent/helpers.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 langroid/agent/junk,sha256=LxfuuW7Cijsg0szAzT81OjWWv1PMNI-6w_-DspVIO2s,339
 langroid/agent/openai_assistant.py,sha256=2rjCZw45ysNBEGNzQM4uf0bTC4KkatGYAWcVcW4xcek,34337
 langroid/agent/special/__init__.py,sha256=gik_Xtm_zV7U9s30Mn8UX3Gyuy4jTjQe9zjiE3HWmEo,1273
-langroid/agent/special/doc_chat_agent.py,sha256=3EICtutRADu8S8v0qO8PGFu3VyqjDY6Gp8xYgNtiNSY,54596
+langroid/agent/special/doc_chat_agent.py,sha256=dqm0Gp11Mfl4hOWN4sUR1uZL-oHEmHzcB6bNN6WFgqw,54784
 langroid/agent/special/lance_doc_chat_agent.py,sha256=s8xoRs0gGaFtDYFUSIRchsgDVbS5Q3C2b2mr3V1Fd-Q,10419
 langroid/agent/special/lance_rag/__init__.py,sha256=QTbs0IVE2ZgDg8JJy1zN97rUUg4uEPH7SLGctFNumk4,174
 langroid/agent/special/lance_rag/critic_agent.py,sha256=OtFuHthKQLkdVkvuZ2m0GNq1qOYLqHkm1pfLRFnSg5c,9548
@@ -91,7 +91,7 @@ langroid/parsing/parse_json.py,sha256=sKrYv9-IUqRFaTJA24_rmfjN1E7dQSrTBrtd1jYDE1
 langroid/parsing/parser.py,sha256=AgtmlVUvrkSG1l7-YZPX8rlldgXjh_HqXAMqpXkBxUo,11746
 langroid/parsing/repo_loader.py,sha256=3GjvPJS6Vf5L6gV2zOU8s-Tf1oq_fZm-IB_RL_7CTsY,29373
 langroid/parsing/routing.py,sha256=-FcnlqldzL4ZoxuDwXjQPNHgBe9F9-F4R6q7b_z9CvI,1232
-langroid/parsing/search.py,sha256=plQtjarB9afGfJLB0CyPXPq3mM4m7kRsfd0_4brziEI,8846
+langroid/parsing/search.py,sha256=0i_r0ESb5HEQfagA2g7_uMQyxYPADWVbdcN9ixZhS4E,8992
 langroid/parsing/spider.py,sha256=Y6y7b86Y2k770LdhxgjVlImBxuuy1V9n8-XQ3QPaG5s,3199
 langroid/parsing/table_loader.py,sha256=qNM4obT_0Y4tjrxNBCNUYjKQ9oETCZ7FbolKBTcz-GM,3410
 langroid/parsing/url_loader.py,sha256=Na2TBlKuQkloZzkE2d7xl6mh9olS3CbpgCsJbJ-xhIA,4472
@@ -137,8 +137,8 @@ langroid/vector_store/meilisearch.py,sha256=6frB7GFWeWmeKzRfLZIvzRjllniZ1cYj3Hmh
 langroid/vector_store/momento.py,sha256=qR-zBF1RKVHQZPZQYW_7g-XpTwr46p8HJuYPCkfJbM4,10534
 langroid/vector_store/qdrant_cloud.py,sha256=3im4Mip0QXLkR6wiqVsjV1QvhSElfxdFSuDKddBDQ-4,188
 langroid/vector_store/qdrantdb.py,sha256=v88lqFkepADvlN6lByUj9I4NEKa9X9lWH16uTPPbYrE,17457
-pyproject.toml,sha256=oocGdj8dqhrarP8c5LeFeOKboZ4WYNzs1YpcKszoJgM,7107
-langroid-0.12.0.dist-info/LICENSE,sha256=EgVbvA6VSYgUlvC3RvPKehSg7MFaxWDsFuzLOsPPfJg,1065
-langroid-0.12.0.dist-info/METADATA,sha256=S-V-w4lhAay08FYPRyJcT7FliitUSaLkL9gQS25luSQ,55259
-langroid-0.12.0.dist-info/WHEEL,sha256=FMvqSimYX_P7y0a7UY-_Mc83r5zkBZsCYPm7Lr0Bsq4,88
-langroid-0.12.0.dist-info/RECORD,,
+pyproject.toml,sha256=g99bgxP-XUiTx-KsdFICVJuV2bB89areQkDRU5sIgmk,7107
+langroid-0.13.0.dist-info/LICENSE,sha256=EgVbvA6VSYgUlvC3RvPKehSg7MFaxWDsFuzLOsPPfJg,1065
+langroid-0.13.0.dist-info/METADATA,sha256=Znhge-Z8nn_L7Lxeh8dWs04d4ejZfj0NCCRutJJSkdg,55259
+langroid-0.13.0.dist-info/WHEEL,sha256=FMvqSimYX_P7y0a7UY-_Mc83r5zkBZsCYPm7Lr0Bsq4,88
+langroid-0.13.0.dist-info/RECORD,,

pyproject.toml

@@ -1,6 +1,6 @@
 [tool.poetry]
 name = "langroid"
-version = "0.12.0"
+version = "0.13.0"
 description = "Harness LLMs with Multi-Agent Programming"
 authors = ["Prasad Chalasani <pchalasani@gmail.com>"]
 readme = "README.md"