fix(rag): reject invalid chunk_size/chunk_overlap in RecursiveCharacterTextSplitter

Fixes #38980

litellm/rag/text_splitters/recursive_character_text_splitter.py

@@ -23,6 +23,15 @@ class RecursiveCharacterTextSplitter:
         chunk_overlap: int = DEFAULT_CHUNK_OVERLAP,
         separators: list[str] | None = None,
     ):
+        if chunk_size <= 0:
+            raise ValueError(f"chunk_size must be positive, got {chunk_size}")
+        if chunk_overlap < 0:
+            raise ValueError(f"chunk_overlap must be non-negative, got {chunk_overlap}")
+        if chunk_overlap >= chunk_size:
+            raise ValueError(
+                "chunk_overlap must be smaller than chunk_size, "
+                f"got chunk_overlap={chunk_overlap}, chunk_size={chunk_size}"
+            )
         self.chunk_size = chunk_size
         self.chunk_overlap = chunk_overlap
         self.separators = separators or ["\n\n", "\n", " ", ""]

tests/unit/rag/text_splitters/test_recursive_character_text_splitter.py

@@ -0,0 +1,59 @@
+from typing import Final
+
+import pytest
+
+from litellm.constants import DEFAULT_CHUNK_OVERLAP, DEFAULT_CHUNK_SIZE
+from litellm.rag.text_splitters.recursive_character_text_splitter import (
+    RecursiveCharacterTextSplitter,
+)
+
+
+@pytest.mark.parametrize(
+    "chunk_size,chunk_overlap",
+    [
+        (100, 100),
+        (100, 150),
+        (1, 1),
+    ],
+)
+def test_chunk_overlap_at_least_chunk_size_is_rejected(
+    chunk_size: int, chunk_overlap: int
+) -> None:
+    with pytest.raises(
+        ValueError, match="chunk_overlap must be smaller than chunk_size"
+    ):
+        RecursiveCharacterTextSplitter(
+            chunk_size=chunk_size, chunk_overlap=chunk_overlap
+        )
+
+
+@pytest.mark.parametrize("chunk_size", [0, -1, -100])
+def test_non_positive_chunk_size_is_rejected(chunk_size: int) -> None:
+    with pytest.raises(ValueError, match="chunk_size must be positive"):
+        RecursiveCharacterTextSplitter(chunk_size=chunk_size)
+
+
+@pytest.mark.parametrize("chunk_overlap", [-1, -100])
+def test_negative_chunk_overlap_is_rejected(chunk_overlap: int) -> None:
+    with pytest.raises(ValueError, match="chunk_overlap must be non-negative"):
+        RecursiveCharacterTextSplitter(chunk_overlap=chunk_overlap)
+
+
+def test_force_split_path_terminates_with_valid_overlap() -> None:
+    splitter: Final = RecursiveCharacterTextSplitter(
+        chunk_size=10, chunk_overlap=2, separators=["\n\n"]
+    )
+    chunks: Final = splitter.split_text("abcdefghijklmnop")
+    assert chunks == ["abcdefghij", "ijklmnop"]
+
+
+def test_valid_overlap_keeps_splitting() -> None:
+    splitter: Final = RecursiveCharacterTextSplitter(chunk_size=10, chunk_overlap=2)
+    chunks: Final = splitter.split_text("abcdefghijklmnop")
+    assert chunks == ["abcdefghij", "ijklmnop"]
+
+
+def test_defaults_construct_without_error() -> None:
+    splitter: Final = RecursiveCharacterTextSplitter()
+    assert splitter.chunk_size == DEFAULT_CHUNK_SIZE
+    assert splitter.chunk_overlap == DEFAULT_CHUNK_OVERLAP