!fix: return a list of elements in stage_for_transformers (#420)

* update stage_for_transformers to return a list of elements

* bump changelog and version

* flag breaking change

* fix last word bug in chunk_by_attention_window

CHANGELOG.md

@@ -1,10 +1,11 @@
-## 0.5.8-dev4
+## 0.5.8-dev5
 
 ### Enhancements
 
 * Update `elements_to_json` to return string when filename is not specified
 * `elements_from_json` may take a string instead of a filename with the `text` kwarg
 * `detect_filetype` now does a final fallback to file extension.
+* Empty tags are now skipped during the depth check for HTML processing.
 
 ### Features
 
@@ -18,6 +19,13 @@
 * Partitioning functions that accept a `text` kwarg no longer raise an error if an empty
   string is passed (and empty list of elements is returned instead).
 * `partition_json` no longer fails if the input is an empty list.
+* Fixed bug in `chunk_by_attention_window` that caused the last word in segments to be cut-off
+  in some cases.
+
+### BREAKING CHANGES
+
+* `stage_for_transformers` now returns a list of elements, making it consistent with other
+  staging bricks
 
 ## 0.5.7
 

test_unstructured/staging/test_huggingface.py

@@ -1,6 +1,6 @@
 import pytest
 
-from unstructured.documents.elements import Text
+from unstructured.documents.elements import Text, Title
 from unstructured.staging import huggingface
 
 
@@ -12,14 +12,23 @@ class MockTokenizer:
 
 
 def test_stage_for_transformers():
-    elements = [Text(text="hello " * 20), Text(text="there " * 20)]
+    title_element = (Title(text="Here is a wonderful story"),)
+    elements = [title_element, Text(text="hello " * 20 + "there " * 20)]
+
     tokenizer = MockTokenizer()
 
-    chunks = huggingface.stage_for_transformers(elements, tokenizer, buffer=10)
+    chunk_elements = huggingface.stage_for_transformers(elements, tokenizer, buffer=10)
 
-    hello_chunk = ("hello " * 10).strip()
-    there_chunk = ("there " * 10).strip()
-    assert chunks == [hello_chunk, hello_chunk, "\n\n" + there_chunk, there_chunk]
+    hello_chunk = Text(("hello " * 10).strip())
+    there_chunk = Text(("there " * 10).strip())
+
+    assert chunk_elements == [
+        title_element,
+        hello_chunk,
+        hello_chunk,
+        there_chunk,
+        there_chunk,
+    ]
 
 
 def test_chunk_by_attention_window():

unstructured/__version__.py

@@ -1 +1 @@
-__version__ = "0.5.8-dev4"  # pragma: no cover
+__version__ = "0.5.8-dev5"  # pragma: no cover

unstructured/staging/huggingface.py

@@ -1,19 +1,32 @@
+from copy import deepcopy
 from typing import Callable, List, Optional
 
 from transformers import PreTrainedTokenizer
 
-from unstructured.documents.elements import Text
+from unstructured.documents.elements import Element, NarrativeText, Text
 
 
 def stage_for_transformers(
     elements: List[Text],
     tokenizer: PreTrainedTokenizer,
     **chunk_kwargs,
-) -> List[str]:
+) -> List[Element]:
     """Stages text elements for transformers pipelines by chunking them into sections that can
     fit into the attention window for the model associated with the tokenizer."""
-    combined_text = "\n\n".join([str(element) for element in elements])
-    return chunk_by_attention_window(combined_text, tokenizer, **chunk_kwargs)
+    chunked_elements: List[Element] = []
+    for element in elements:
+        # NOTE(robinson) - Only chunk potentially lengthy text. Shorter text (like titles)
+        # should already fit into the attention window just fine.
+        if isinstance(element, (NarrativeText, Text)):
+            chunked_text = chunk_by_attention_window(element.text, tokenizer, **chunk_kwargs)
+            for chunk in chunked_text:
+                _chunk_element = deepcopy(element)
+                _chunk_element.text = chunk
+                chunked_elements.append(_chunk_element)
+        else:
+            chunked_elements.append(element)
+
+    return chunked_elements
 
 
 def chunk_by_attention_window(
@@ -68,8 +81,8 @@ def chunk_by_attention_window(
                 f"error is: \n\n{segment}",
             )
 
-        if chunk_size + num_tokens > max_chunk_size or i == (num_splits - 1):
-            chunks.append(chunk_text)
+        if chunk_size + num_tokens > max_chunk_size:
+            chunks.append(chunk_text + chunk_separator.strip())
             chunk_text = ""
             chunk_size = 0
 
@@ -79,4 +92,7 @@ def chunk_by_attention_window(
         chunk_text += segment
         chunk_size += num_tokens
 
+        if i == (num_splits - 1) and len(chunk_text) > 0:
+            chunks.append(chunk_text)
+
     return chunks