x

libs/langchain_v1/langchain/chains/summarization.py

@@ -1,21 +1,50 @@
-"""Inline Summarization Chain"""
+"""Inline Summarization Chain
+
+TODO(Eugene): see if we can add annotations / citations.
+"""
+
+from __future__ import annotations
 
 from typing import NotRequired, cast
-from typing import TypedDict
-
-from langgraph.graph import StateGraph, END
-from langgraph.pregel import Pregel
 
 from langchain_core.documents import Document
 from langchain_core.language_models.chat_models import BaseChatModel
-from langchain_core.messages import MessageLikeRepresentation, AIMessage
+from langchain_core.messages import AIMessage, MessageLikeRepresentation
+from langgraph.graph import END, StateGraph
+from langgraph.pregel import Pregel
+from typing_extensions import TypedDict
+
+from langchain._internal.utils import RunnableCallable
 
 
 class InlineSummarizationState(TypedDict):
     """State for inline summarization."""
 
     documents: list[Document]
+    """List of documents to summarize."""
     summary: NotRequired[str]
+    """Summary of the documents, available after summarization."""
+
+
+class InputSchema(TypedDict):
+    """Input for the inline summarization chain."""
+
+    documents: list[Document]
+    """List of documents to summarize."""
+
+
+class OutputSchema(TypedDict):
+    """Output of the inline summarization chain."""
+
+    summary: str
+    """Summary of the documents."""
+
+
+class SummarizationNodeUpdate(TypedDict):
+    """Update for the summarization node."""
+
+    summary: str
+    """Summary of the documents."""
 
 
 class InlineSummarizer:
@@ -54,28 +83,40 @@ class InlineSummarizer:
             },
         ]
 
-    def _summarize_node(self, state: InlineSummarizationState) -> TypedDict(
-        "Update", {"summary": str}
-    ):
+    def create_summarization_node(
+        self,
+    ) -> RunnableCallable[InlineSummarizationState, SummarizationNodeUpdate]:
+        """Creates a node for inline summarization."""
+
+        def _summarize_node(state: InlineSummarizationState) -> SummarizationNodeUpdate:
             """Builds a LangGraph for inline summarization."""
             prompt = self._get_prompt(state)
-        response = cast(AIMessage, self.model.invoke(prompt))
+            response = cast("AIMessage", self.model.invoke(prompt))
             return {"summary": response.text()}
 
-    async def _asummarize_node(self, state: InlineSummarizationState) -> TypedDict(
-        "Update", {"summary": str}
-    ):
+        async def _asummarize_node(
+            state: InlineSummarizationState,
+        ) -> SummarizationNodeUpdate:
             """Asynchronous version of the summarize node."""
             prompt = self._get_prompt(state)
-        response = cast(AIMessage, await self.model.ainvoke(prompt))
+            response = cast("AIMessage", await self.model.ainvoke(prompt))
             return {"summary": response.text()}
 
+        return RunnableCallable[InlineSummarizationState, SummarizationNodeUpdate](
+            _summarize_node,
+            _asummarize_node,
+        )
+
     def build(self) -> Pregel:
         """Builds the LangGraph for inline summarization."""
-        builder = StateGraph(InlineSummarizationState)
-        builder.add_node("summarize_inline", self._summarize_node)
-        builder.set_entry_point("summarize_inline")
-        builder.add_edge("summarize_inline", END)
+        builder = StateGraph(
+            InlineSummarizationState,
+            input_schema=InputSchema,
+            output_schema=OutputSchema,
+        )
+        builder.add_node("summarize", self.create_summarization_node())
+        builder.set_entry_point("summarize")
+        builder.add_edge("summarize", END)
         return builder