Fix: overlap cannot be properly applied (#12828)

### What problem does this PR solve?

Overlap cannot be properly applied.

### Type of change

- [x] Bug Fix (non-breaking change which fixes an issue)

common/float_utils.py

@@ -14,6 +14,7 @@
 #  limitations under the License.
 #
 
+
 def get_float(v):
     """
     Convert a value to float, handling None and exceptions gracefully.
@@ -39,8 +40,19 @@ def get_float(v):
         42.0
     """
     if v is None:
-        return float('-inf')
+        return float("-inf")
     try:
         return float(v)
     except Exception:
-        return float('-inf')
+        return float("-inf")
+
+
+def normalize_overlapped_percent(overlapped_percent):
+    try:
+        value = float(overlapped_percent)
+    except (TypeError, ValueError):
+        return 0
+    if 0 < value < 1:
+        value *= 100
+    value = int(value)
+    return max(0, min(value, 90))

rag/app/naive.py

@@ -38,6 +38,7 @@ from deepdoc.parser.figure_parser import VisionFigureParser, vision_figure_parse
 from deepdoc.parser.pdf_parser import PlainParser, VisionParser
 from deepdoc.parser.docling_parser import DoclingParser
 from deepdoc.parser.tcadp_parser import TCADPParser
+from common.float_utils import normalize_overlapped_percent
 from common.parser_config_utils import normalize_layout_recognizer
 from rag.nlp import (
     concat_img,
@@ -983,12 +984,11 @@ def chunk(filename, binary=None, from_page=0, to_page=100000, lang="Chinese", ca
         raise NotImplementedError("file type not supported yet(pdf, xlsx, doc, docx, txt supported)")
 
     st = timer()
+    overlapped_percent = normalize_overlapped_percent(parser_config.get("overlapped_percent", 0))
     if is_markdown:
         merged_chunks = []
         merged_images = []
         chunk_limit = max(0, int(parser_config.get("chunk_token_num", 128)))
-        overlapped_percent = int(parser_config.get("overlapped_percent", 0))
-        overlapped_percent = max(0, min(overlapped_percent, 90))
 
         current_text = ""
         current_tokens = 0
@@ -1037,10 +1037,10 @@ def chunk(filename, binary=None, from_page=0, to_page=100000, lang="Chinese", ca
                 section_images = None
 
         if section_images:
-            chunks, images = naive_merge_with_images(sections, section_images, int(parser_config.get("chunk_token_num", 128)), parser_config.get("delimiter", "\n!?。；！？"))
+            chunks, images = naive_merge_with_images(sections, section_images, int(parser_config.get("chunk_token_num", 128)), parser_config.get("delimiter", "\n!?。；！？"), overlapped_percent)
             res.extend(tokenize_chunks_with_images(chunks, doc, is_english, images, child_delimiters_pattern=child_deli))
         else:
-            chunks = naive_merge(sections, int(parser_config.get("chunk_token_num", 128)), parser_config.get("delimiter", "\n!?。；！？"))
+            chunks = naive_merge(sections, int(parser_config.get("chunk_token_num", 128)), parser_config.get("delimiter", "\n!?。；！？"), overlapped_percent)
 
             res.extend(tokenize_chunks(chunks, doc, is_english, pdf_parser, child_delimiters_pattern=child_deli))
 

rag/flow/splitter/splitter.py

@@ -23,6 +23,7 @@ from rag.utils.base64_image import id2image, image2id
 from deepdoc.parser.pdf_parser import RAGFlowPdfParser
 from rag.flow.base import ProcessBase, ProcessParamBase
 from rag.flow.splitter.schema import SplitterFromUpstream
+from common.float_utils import normalize_overlapped_percent
 from rag.nlp import attach_media_context, naive_merge, naive_merge_with_images
 from common import settings
 
@@ -68,6 +69,7 @@ class Splitter(ProcessBase):
 
         self.set_output("output_format", "chunks")
         self.callback(random.randint(1, 5) / 100.0, "Start to split into chunks.")
+        overlapped_percent = normalize_overlapped_percent(self._param.overlapped_percent)
         if from_upstream.output_format in ["markdown", "text", "html"]:
             if from_upstream.output_format == "markdown":
                 payload = from_upstream.markdown_result
@@ -83,7 +85,7 @@ class Splitter(ProcessBase):
                 payload,
                 self._param.chunk_token_size,
                 deli,
-                self._param.overlapped_percent,
+                overlapped_percent,
             )
             if custom_pattern:
                 docs = []
@@ -129,7 +131,7 @@ class Splitter(ProcessBase):
             section_images,
             self._param.chunk_token_size,
             deli,
-            self._param.overlapped_percent,
+            overlapped_percent,
         )
         cks = [
             {