RAGFlow is an agent-native database: a Retrieval-Augmented Generation (RAG) engine that grew into context infrastructure.
Go agent-harnessagent-native-databaseagentic-aiagentic-retrievalagentic-searchaiai-agentscontext-enginecontext-engineeringcontext-managementharness-engineeringknowledge-compilationragretrieval-augmented-generationretrieval-harnesssearch-harness
fix: separate raptor node target from output limit (#17792)
buua4360f04f4c3b970
diff --git a/rag/advanced_rag/knowlege_compile/raptor.py b/rag/advanced_rag/knowlege_compile/raptor.pyindex 587eb1f24b..b9c33d7fbe 100644--- a/rag/advanced_rag/knowlege_compile/raptor.py+++ b/rag/advanced_rag/knowlege_compile/raptor.py@@ -32,6 +32,8 @@ from rag.graphrag.utils import ()from common.misc_utils import thread_pool_exec+from ._common import knowledge_compile_gen_conf+class RecursiveAbstractiveProcessing4TreeOrganizedRetrieval:"""Build RAPTOR summary layers with the classic or Psi tree strategy."""@@ -221,10 +223,18 @@ class RecursiveAbstractiveProcessing4TreeOrganizedRetrieval:[{"role": "user",- "content": "Beside the summarization, give a title at the first line of your summarization. Must be in the same language as the paragraphs.",+ "content": (+ "Beside the summarization, give a title at the first line of your summarization. "+ "Must be in the same language as the paragraphs. "+ f"Keep the summary concise and target approximately {self._max_token} tokens."+ ),}],- {"max_tokens": max(self._max_token, 512)}, # fix issue: #10235+ # ``max_token`` is the target size of the generated node,+ # not the provider's per-request output ceiling. Keep the+ # provider budget independent so reasoning tokens cannot+ # consume the node-size setting and truncate the summary.+ knowledge_compile_gen_conf(self._llm_model),)cnt = re.sub("(······\n由于长度的原因,回答被截断了,要继续吗?|For the content length reason, it stopped, continue?)",