Set repeat_table_header=False to match docling serve

2026-03-05 13:02:18 +02:00 · 2026-03-05 13:02:18 +02:00 · d460d553f3
commit d460d553f3
parent 30ebd6694b
2 changed files with 16166 additions and 16344 deletions
--- a/haiku_rag_slim/haiku/rag/chunkers/docling_local.py
+++ b/haiku_rag_slim/haiku/rag/chunkers/docling_local.py
@ -83,6 +83,9 @@ class DoclingLocalChunker(DocumentChunker):
                tokenizer=tokenizer,
                merge_peers=config.processing.chunking_merge_peers,
                serializer_provider=serializer_provider,
+                # Context expansion reconstructs full tables, so repeating
+                # headers in each chunk is redundant and inflates chunk count.
+                repeat_table_header=False,
            )
        elif self.chunker_type == "hierarchical":
            serializer_provider = _create_markdown_serializer_provider(
--- a/tests/cassettes/test_chunker/test_local_and_serve_chunkers_produce_same_output.yaml
+++ b/tests/cassettes/test_chunker/test_local_and_serve_chunkers_produce_same_output.yaml