fix(dataprep): guard smart_chunk_text against stride >= chunk_size (#7126)

RawTextDataLoader.smart_chunk_text takes chunk_size and stride as its own
arguments, so a direct call with stride >= chunk_size bypasses the
constructor validation. In that case `start_idx += chunk_size - stride` is
non-positive, so start_idx never advances past the first window and the
chunking loop never terminates (hangs).

Re-add the chunk_size/stride guard at the top of smart_chunk_text so direct
callers fail fast with a clear ValueError. The constructor keeps its own
guard for the internal callers (defense in depth). Add a regression test
that calls smart_chunk_text directly with stride == chunk_size and
stride > chunk_size and asserts it raises instead of hanging.

Signed-off-by: Anas Khan <83116240+anxkhn@users.noreply.github.com>

tests/test_raw_text.py

@@ -141,6 +141,26 @@ def test_raw_text_loader():
         except ValueError as e:
             assert "stride" in str(e) and "chunk_size" in str(e)
 
+        # smart_chunk_text validation: called directly, chunk_size/stride are its own
+        # arguments and bypass the constructor guard, so it must guard itself or an
+        # invalid stride makes `start_idx += chunk_size - stride` non-positive and the
+        # chunking loop never terminates (hangs).
+        long_text = "This is a test file for raw text training. " * 10
+        valid_chunks = loader.smart_chunk_text(long_text, chunk_size = 5, stride = 2)
+        assert len(valid_chunks) > 0, "Valid stride should produce chunks"
+
+        try:
+            loader.smart_chunk_text(long_text, chunk_size = 5, stride = 5)
+            assert False, "Should raise ValueError for stride == chunk_size"
+        except ValueError as e:
+            assert "stride" in str(e) and "chunk_size" in str(e)
+
+        try:
+            loader.smart_chunk_text(long_text, chunk_size = 5, stride = 10)
+            assert False, "Should raise ValueError for stride > chunk_size"
+        except ValueError as e:
+            assert "stride" in str(e) and "chunk_size" in str(e)
+
         # Preprocessor.
         preprocessor = TextPreprocessor()
         clean_text = preprocessor.clean_text("  messy   text  \n\n\n  ")

unsloth/dataprep/raw_text.py

@@ -132,6 +132,13 @@ class RawTextDataLoader:
         3. Maintains context with stride overlap
         4. Returns tokenized chunks directly (more efficient) or text chunks
         """
+        if chunk_size <= 0:
+            raise ValueError(f"chunk_size must be positive, got {chunk_size}")
+        if stride >= chunk_size:
+            raise ValueError(
+                f"stride ({stride}) must be smaller than chunk_size ({chunk_size}) to progress the chunking loop"
+            )
+
         # Tokenize the whole text once for accurate token counts
         tokenized = self.tokenizer(text, return_tensors = "pt", add_special_tokens = False)
         tokens = tokenized["input_ids"]