fix: add final answer check

libs/ktem/ktem/index/file/knet/pipelines.py

@@ -101,12 +101,12 @@ class KnetRetrievalPipeline(BaseFileIndexRetriever):
             chunks = yaml.safe_load(response.content)
             for chunk in chunks:
                 metadata = chunk["node"]["metadata"]
+                content_type = metadata.pop("content_type", "")
+
                 metadata["page_label"] = metadata.get(
                     "pageIdx", metadata.get("parentPageIdx", "")
                 )
-                metadata["type"] = metadata_translation.get(
-                    metadata.pop("content_type", ""), ""
-                )
+                metadata["type"] = metadata_translation.get(content_type, content_type)
                 metadata["file_name"] = metadata.pop("company_name", "")
 
                 # load image from returned path

libs/ktem/ktem/reasoning/simple.py

@@ -205,6 +205,15 @@ DEFAULT_REWRITE_PROMPT = (
 
 CONTEXT_RELEVANT_WARNING_SCORE = 0.7
 
+FINAL_ANSWER_CONTENT_TYPES = [
+    "long_term_memory_index",
+    "tagged",
+    "number_of_page_nodes",
+    "document_page_info",
+    "template_based",
+    "chat_completion",
+]
+
 
 class AnswerWithContextPipeline(BaseComponent):
     """Answer the question based on the evidence
@@ -696,6 +705,15 @@ class FullQAPipeline(BaseReasoning):
             if without_citation:
                 yield from without_citation
 
+    def find_final_answer(self, docs):
+        final_answer = None
+        for doc in docs:
+            if doc.metadata.get("type") in FINAL_ANSWER_CONTENT_TYPES:
+                final_answer = doc
+                break
+
+        return final_answer
+
     async def ainvoke(  # type: ignore
         self, message: str, conv_id: str, history: list, **kwargs  # type: ignore
     ) -> Document:  # type: ignore
@@ -748,6 +766,16 @@ class FullQAPipeline(BaseReasoning):
         print(f"Got {len(docs)} retrieved documents")
         yield from infos
 
+        # check if evidences contain final answer
+        final_answer = self.find_final_answer(docs)
+        if final_answer:
+            yield Document(
+                channel="chat",
+                content=final_answer.text,
+            )
+            print("No LLM generation because final answer found in evidences")
+            return final_answer.text
+
         evidence_mode, evidence = self.evidence_pipeline(docs).content
 
         def generate_relevant_scores():