Skip to content

RAGFlow is an agent-native database: a Retrieval-Augmented Generation (RAG) engine that grew into context infrastructure.

fix: use updated RAPTOR options for skill generation (#17859)

buua436502921159a06
diff --git a/rag/svr/task_executor_refactor/dataset_skill_generator.py b/rag/svr/task_executor_refactor/dataset_skill_generator.py
index c447d1fb03..77ce8e3fff 100644
--- a/rag/svr/task_executor_refactor/dataset_skill_generator.py
+++ b/rag/svr/task_executor_refactor/dataset_skill_generator.py
@@ -366,9 +366,10 @@ async def run_corpus2skill(
llm_model=chat_mdl,
embd_model=embedding_model,
prompt="Please write a concise summary of the following texts:\n{cluster_content}",
- max_token=256,
- threshold=0.1,
+ max_token=512,
max_errors=3,
+ clustering_threshold=0.3,
+ clustering_ratio=0.5,
)
# ---- Phase 1: per-doc summaries.