RecursiveCharacterTextSplitter(chunk_size=100) alone raises "Got a larger chunk overlap (200)"
Symptom: RecursiveCharacterTextSplitter(chunk_size=100), CharacterTextSplitter(chunk_size=100) or .from_tiktoken_encoder(chunk_size=50) raise ValueError: Got a larger chunk overlap (200) than chunk size (100), should be smaller, even though you never passed chunk_overlap.
Fix: always pass chunk_overlap explicitly and smaller than chunk_size, e.g. RecursiveCharacterTextSplitter(chunk_size=100, chunk_overlap=10). The base-class defaults are chunk_size=4000 and chunk_overlap=200, so any chunk_size <= 200 fails unless overlap is set.
With tiktoken the traceback goes through from_tiktoken_encoder -> cls(...) -> TextSplitter.__init__, so it looks like a tokenizer problem but is not.
Output: both classes with {chunk_size: 100} raise the error above; with chunk_overlap=0 they construct fine. inspect.signature(TextSplitter.__init__) shows chunk_size=4000, chunk_overlap=200.
Also: from langchain.text_splitter import ... gives ModuleNotFoundError: No module named langchain.text_splitter on langchain 1.5.0; import from langchain_text_splitters.
Versions: langchain-text-splitters 1.1.3, langchain 1.5.0, Python 3.12.