langroid 0.45.1__tar.gz → 0.45.3__tar.gz

{langroid-0.45.1 → langroid-0.45.3}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: langroid
-Version: 0.45.1
+Version: 0.45.3
 Summary: Harness LLMs with Multi-Agent Programming
 Author-email: Prasad Chalasani <pchalasani@gmail.com>
 License: MIT

{langroid-0.45.1 → langroid-0.45.3}/langroid/agent/base.py

@@ -90,6 +90,7 @@ class AgentConfig(BaseSettings):
     parsing: Optional[ParsingConfig] = ParsingConfig()
     prompts: Optional[PromptsConfig] = PromptsConfig()
     show_stats: bool = True  # show token usage/cost stats?
+    hide_agent_response: bool = True  # hide agent response?
     add_to_registry: bool = True  # register agent in ObjectRegistry?
     respond_tools_only: bool = False  # respond only to tool messages (not plain text)?
     # allow multiple tool messages in a single response?
@@ -460,6 +461,28 @@ class Agent(ABC):
             recipient=recipient,
         )
 
+    def render_agent_response(
+        self,
+        results: Optional[str | OrderedDict[str, str] | ChatDocument],
+    ) -> None:
+        """
+        Render the response from the agent, typically from tool-handling.
+        Args:
+            results: results from tool-handling, which may be a string,
+                a dict of tool results, or a ChatDocument.
+        """
+        if self.config.hide_agent_response or results is None:
+            return
+        if isinstance(results, str):
+            results_str = results
+        elif isinstance(results, ChatDocument):
+            results_str = results.content
+        elif isinstance(results, dict):
+            results_str = json.dumps(results, indent=2)
+        if not settings.quiet:
+            console.print(f"[red]{self.indent}", end="")
+            print(f"[red]Agent: {escape(results_str)}")
+
     def _agent_response_final(
         self,
         msg: Optional[str | ChatDocument],
@@ -477,8 +500,7 @@ class Agent(ABC):
         elif isinstance(results, dict):
             results_str = json.dumps(results, indent=2)
         if not settings.quiet:
-            console.print(f"[red]{self.indent}", end="")
-            print(f"[red]Agent: {escape(results_str)}")
+            self.render_agent_response(results)
         maybe_json = len(extract_top_level_json(results_str)) > 0
         self.callbacks.show_agent_response(
             content=results_str,
@@ -1341,8 +1363,7 @@ class Agent(ABC):
 
         has_orch = any(isinstance(t, ORCHESTRATION_TOOLS) for t in tools)
         if has_orch and len(tools) > 1:
-            err_str = "ERROR: Use ONE tool at a time!"
-            return [err_str for _ in tools]
+            return ["ERROR: Use ONE tool at a time!"] * len(tools)
 
         return []
 
@@ -1477,8 +1498,6 @@ class Agent(ABC):
             # as a response to the tool message even though the tool was not intended
             # for this agent.
             return None
-        if len(tools) > 1 and not self.config.allow_multiple_tools:
-            return self.to_ChatDocument("ERROR: Use ONE tool at a time!")
         if len(tools) == 0:
             fallback_result = self.handle_message_fallback(msg)
             if fallback_result is None:
@@ -1487,10 +1506,14 @@ class Agent(ABC):
                 fallback_result,
                 chat_doc=msg if isinstance(msg, ChatDocument) else None,
             )
-        chat_doc = msg if isinstance(msg, ChatDocument) else None
 
-        results = self._get_multiple_orch_tool_errs(tools)
+        results: List[str | ChatDocument | None] = []
+        if len(tools) > 1 and not self.config.allow_multiple_tools:
+            results = ["ERROR: Use ONE tool at a time!"] * len(tools)
+        if not results:
+            results = self._get_multiple_orch_tool_errs(tools)
         if not results:
+            chat_doc = msg if isinstance(msg, ChatDocument) else None
             results = [self.handle_tool_message(t, chat_doc=chat_doc) for t in tools]
             # if there's a solitary ChatDocument|str result, return it as is
             if len(results) == 1 and isinstance(results[0], (str, ChatDocument)):

{langroid-0.45.1 → langroid-0.45.3}/langroid/agent/chat_agent.py

@@ -85,6 +85,8 @@ class ChatAgentConfig(AgentConfig):
             enabled when such tool calls are not desired.
         output_format_include_defaults: Whether to include fields with default arguments
             in the output schema
+        full_citations: Whether to show source reference citation + content for each
+            citation, or just the main reference citation.
     """
 
     system_message: str = "You are a helpful assistant."
@@ -101,6 +103,7 @@ class ChatAgentConfig(AgentConfig):
     instructions_output_format: bool = True
     output_format_include_defaults: bool = True
     use_tools_on_output_format: bool = True
+    full_citations: bool = True  # show source + content for each citation?
 
     def _set_fn_or_tools(self, fn_available: bool) -> None:
         """
@@ -1854,14 +1857,15 @@ class ChatAgent(Agent):
             # we won't have citations yet, so we're done
             return
         if response.metadata.has_citation:
+            citation = (
+                response.metadata.source_content
+                if self.config.full_citations
+                else response.metadata.source
+            )
             if not settings.quiet:
-                print(
-                    "[grey37]SOURCES:\n"
-                    + escape(response.metadata.source)
-                    + "[/grey37]"
-                )
+                print("[grey37]SOURCES:\n" + escape(citation) + "[/grey37]")
             self.callbacks.show_llm_response(
-                content=str(response.metadata.source),
+                content=str(citation),
                 is_tool=False,
                 cached=False,
                 language="text",

{langroid-0.45.1 → langroid-0.45.3}/langroid/agent/special/doc_chat_agent.py

@@ -863,12 +863,13 @@ class DocChatAgent(ChatAgent):
         # extract references like [^2], [^3], etc. from the final answer
         citations = extract_markdown_references(final_answer)
         # format the cited references as a string suitable for markdown footnote
-        citations_str = format_cited_references(citations, passages)
+        full_citations_str, citations_str = format_cited_references(citations, passages)
 
         return ChatDocument(
             content=final_answer,  # does not contain citations
             metadata=ChatDocMetaData(
-                source=citations_str,  # only the citations
+                source=citations_str,  # only the reference headers
+                source_content=full_citations_str,  # reference + content
                 sender=Entity.LLM,
                 has_citation=len(citations) > 0,
                 cached=getattr(answer_doc.metadata, "cached", False),

{langroid-0.45.1 → langroid-0.45.3}/langroid/mytypes.py

@@ -43,7 +43,8 @@ class Entity(str, Enum):
 class DocMetaData(BaseModel):
     """Metadata for a document."""
 
-    source: str = "context"
+    source: str = "context"  # just reference
+    source_content: str = "context"  # reference and content
     is_chunk: bool = False  # if it is a chunk, don't split
     id: str = Field(default_factory=lambda: str(uuid4()))
     window_ids: List[str] = []  # for RAG: ids of chunks around this one

{langroid-0.45.1 → langroid-0.45.3}/langroid/parsing/document_parser.py

@@ -404,8 +404,8 @@ class DocumentParser(Parser):
             # that it needs to be combined with the next chunk.
             while len(split) > self.config.chunk_size:
                 # pretty formatting of pages (e.g. 1-3, 4, 5-7)
-                p_0 = int(pages[0])
-                p_n = int(pages[-1])
+                p_0 = int(pages[0]) - self.config.page_number_offset
+                p_n = int(pages[-1]) - self.config.page_number_offset
                 page_str = f"pages {p_0}-{p_n}" if p_0 != p_n else f"page {p_0}"
                 text = self.tokenizer.decode(split[: self.config.chunk_size])
                 docs.append(
@@ -426,13 +426,15 @@ class DocumentParser(Parser):
         # since it's already included in the prior chunk;
         # the only exception is if there have been no chunks so far.
         if len(split) > self.config.overlap or n_chunks == 0:
-            pg = "-".join([pages[0], pages[-1]])
+            p_0 = int(pages[0]) - self.config.page_number_offset
+            p_n = int(pages[-1]) - self.config.page_number_offset
+            page_str = f"pages {p_0}-{p_n}" if p_0 != p_n else f"page {p_0}"
             text = self.tokenizer.decode(split[: self.config.chunk_size])
             docs.append(
                 Document(
                     content=text,
                     metadata=DocMetaData(
-                        source=f"{self.source} pages {pg}",
+                        source=f"{self.source} {page_str}",
                         is_chunk=True,
                         id=common_id,
                     ),
@@ -1361,6 +1363,10 @@ class GeminiPdfParser(DocumentParser):
 
 
 class MarkerPdfParser(DocumentParser):
+    """
+    Parse PDF files using the `marker` library: https://github.com/VikParuchuri/marker
+    """
+
     DEFAULT_CONFIG = {"paginate_output": True, "output_format": "markdown"}
 
     def __init__(self, source: Union[str, bytes], config: ParsingConfig):

{langroid-0.45.1 → langroid-0.45.3}/langroid/parsing/parser.py

@@ -103,6 +103,9 @@ class ParsingConfig(BaseSettings):
     chunk_size: int = 200  # aim for this many tokens per chunk
     overlap: int = 50  # overlap between chunks
     max_chunks: int = 10_000
+    # offset to subtract from page numbers:
+    # e.g. if physical page 12 is displayed as page 1, set page_number_offset = 11
+    page_number_offset: int = 0
     # aim to have at least this many chars per chunk when truncating due to punctuation
     min_chunk_chars: int = 350
     discard_chunk_chars: int = 5  # discard chunks with fewer than this many chars

{langroid-0.45.1 → langroid-0.45.3}/langroid/utils/output/citations.py

@@ -1,4 +1,4 @@
-from typing import List
+from typing import List, Tuple
 
 from langroid.mytypes import Document
 
@@ -66,7 +66,9 @@ def format_footnote_text(content: str, width: int = 0) -> str:
     return "\n".join(output_lines)
 
 
-def format_cited_references(citations: List[int], passages: list[Document]) -> str:
+def format_cited_references(
+    citations: List[int], passages: list[Document]
+) -> Tuple[str, str]:
     """
     Given a list of (integer) citations, and a list of passages, return a string
     that can be added as a footer to the main text, to show sources cited.
@@ -76,16 +78,24 @@ def format_cited_references(citations: List[int], passages: list[Document]) -> s
         passages (list[Document]): list of passages (Document objects)
 
     Returns:
-        str: formatted string of citations for footnote in markdown
+        str: formatted string of FULL citations (i.e. reference AND content)
+            for footnote in markdown;
+        str: formatted string of BRIEF citations (i.e. reference only)
+            for footnote in markdown.
     """
     citations_str = ""
+    full_citations_str = ""
     if len(citations) > 0:
         # append [i] source, content for each citation
-        citations_str = "\n".join(
+        full_citations_str = "\n".join(
             [
                 f"[^{c}] {passages[c-1].metadata.source}"
                 f"\n{format_footnote_text(passages[c-1].content)}"
                 for c in citations
             ]
         )
-    return citations_str
+        # append [i] source for each citation
+        citations_str = "\n".join(
+            [f"[^{c}] {passages[c-1].metadata.source}" for c in citations]
+        )
+    return full_citations_str, citations_str

{langroid-0.45.1 → langroid-0.45.3}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "langroid"
-version = "0.45.1"
+version = "0.45.3"
 authors = [
     {name = "Prasad Chalasani", email = "pchalasani@gmail.com"},
 ]