build(deps): bump unstructured_inference to 0.5.0

.github/workflows/ci.yml

@@ -103,7 +103,6 @@ jobs:
     - name: Test
       run: |
         source .venv/bin/activate
-        make install-detectron2
         sudo apt-get update
         sudo apt-get install -y libmagic-dev poppler-utils libreoffice pandoc
         sudo add-apt-repository -y ppa:alex-p/tesseract-ocr5
@@ -144,7 +143,6 @@ jobs:
         DISCORD_TOKEN: ${{ secrets.DISCORD_TOKEN }}
       run: |
         source .venv/bin/activate
-        make install-detectron2
         sudo apt-get update
         sudo apt-get install -y libmagic-dev poppler-utils libreoffice pandoc
         sudo add-apt-repository -y ppa:alex-p/tesseract-ocr5

CHANGELOG.md

@@ -1,3 +1,15 @@
+## 0.7.0-dev0
+
+### Enhancements
+
+* Installing `detectron2` from source is no longer required when using the `local-inference` extra.
+
+### Features
+
+### Fixes
+
+* Better handling of the output order for multicolumn documents when using the `"hi_res"` strategy.
+
 ## 0.6.7
 
 ### Enhancements

Dockerfile

@@ -101,8 +101,7 @@ RUN python3.8 -m pip install pip==${PIP_VERSION} && \
   pip install --no-cache -r requirements/ingest-slack.txt && \
   pip install --no-cache -r requirements/ingest-wikipedia.txt && \
   pip install --no-cache -r requirements/local-inference.txt && \
-  scl enable devtoolset-9 bash && \
-  pip install --no-cache "detectron2@git+https://github.com/facebookresearch/detectron2.git@e2ce8dc#egg=detectron2"
+  scl enable devtoolset-9 bash
 
 COPY example-docs example-docs
 COPY unstructured unstructured

Makefile

@@ -103,7 +103,7 @@ install-detectron2: install-tensorboard
 
 ## install-local-inference: installs requirements for local inference
 .PHONY: install-local-inference
-install-local-inference: install install-unstructured-inference install-detectron2
+install-local-inference: install install-unstructured-inference
 
 ## pip-compile:             compiles all base/dev/test requirements
 .PHONY: pip-compile

docker/ubuntu-22/Dockerfile

@@ -16,7 +16,6 @@ SHELL ["/bin/bash", "-c"]
 RUN source ~/.bashrc && pyenv virtualenv 3.8.15 unstructured && \
     source ~/.pyenv/versions/unstructured/bin/activate && \
     make install-ci && \
-    make install-detectron2 && \
     make install-ingest-s3 && \
     make install-ingest-azure && \
     make install-ingest-github && \

docs/source/installing.rst

@@ -17,8 +17,7 @@ installation.
 	* ``libreoffice`` (MS Office docs)
 	* ``pandocs`` (EPUBs, RTFs and Open Office docs)
 
-* If you are parsing PDFs, run the following to install the ``detectron2`` model, which ``unstructured`` uses for layout detection:
-	* ``pip install "detectron2@git+https://github.com/facebookresearch/detectron2.git@e2ce8dc#egg=detectron2"``
+* Follow the instructions `here <https://github.com/Unstructured-IO/unstructured-inference#detectron2>`_ to install ``detectron2``. This is required if you would like to use custom models from the `LayoutParser Model Zoo <https://github.com/Unstructured-IO/unstructured-inference#using-models-from-the-layoutparser-model-zoo>`_.
 
 At this point, you should be able to run the following code:
 

requirements/base.txt

@@ -6,11 +6,11 @@
 #
 anyio==3.6.2
     # via httpcore
-argilla==1.6.0
+argilla==1.7.0
     # via unstructured (setup.py)
 backoff==2.2.1
     # via argilla
-certifi==2022.12.7
+certifi==2023.5.7
     # via
     #   httpcore
     #   httpx
@@ -23,7 +23,9 @@ charset-normalizer==3.1.0
     #   pdfminer-six
     #   requests
 click==8.1.3
-    # via nltk
+    # via
+    #   nltk
+    #   typer
 commonmark==0.9.1
     # via rich
 cryptography==40.0.2
@@ -113,14 +115,19 @@ sniffio==1.3.0
     #   anyio
     #   httpcore
     #   httpx
+tabulate==0.9.0
+    # via unstructured (setup.py)
 tqdm==4.65.0
     # via
     #   argilla
     #   nltk
+typer==0.9.0
+    # via argilla
 typing-extensions==4.5.0
     # via
     #   pydantic
     #   rich
+    #   typer
 urllib3==2.0.2
     # via requests
 wrapt==1.14.1

requirements/local-inference.txt

@@ -7,14 +7,12 @@
 antlr4-python3-runtime==4.9.3
     # via omegaconf
 anyio==3.6.2
-    # via
-    #   httpcore
-    #   starlette
-argilla==1.6.0
+    # via httpcore
+argilla==1.7.0
     # via unstructured (setup.py)
 backoff==2.2.1
     # via argilla
-certifi==2022.12.7
+certifi==2023.5.7
     # via
     #   httpcore
     #   httpx
@@ -29,7 +27,7 @@ charset-normalizer==3.1.0
 click==8.1.3
     # via
     #   nltk
-    #   uvicorn
+    #   typer
 coloredlogs==15.0.1
     # via onnxruntime
 commonmark==0.9.1
@@ -46,23 +44,19 @@ effdet==0.3.0
     # via layoutparser
 et-xmlfile==1.1.0
     # via openpyxl
-fastapi==0.95.1
-    # via unstructured-inference
 filelock==3.12.0
     # via
     #   huggingface-hub
     #   torch
     #   transformers
-flatbuffers==23.3.3
+flatbuffers==23.5.9
     # via onnxruntime
-fonttools==4.39.3
+fonttools==4.39.4
     # via matplotlib
-fsspec==2023.4.0
+fsspec==2023.5.0
     # via huggingface-hub
 h11==0.14.0
-    # via
-    #   httpcore
-    #   uvicorn
+    # via httpcore
 httpcore==0.16.3
     # via httpx
 httpx==0.23.3
@@ -172,16 +166,14 @@ pillow==9.5.0
     #   unstructured (setup.py)
 portalocker==2.7.0
     # via iopath
-protobuf==4.22.4
+protobuf==4.23.1
     # via onnxruntime
 pycocotools==2.0.6
     # via effdet
 pycparser==2.21
     # via cffi
 pydantic==1.10.7
-    # via
-    #   argilla
-    #   fastapi
+    # via argilla
 pygments==2.15.1
     # via rich
 pypandoc==1.11
@@ -225,6 +217,8 @@ rfc3986[idna2008]==1.5.0
     # via httpx
 rich==13.0.1
     # via argilla
+safetensors==0.3.1
+    # via timm
 scipy==1.10.1
     # via layoutparser
 six==1.16.0
@@ -234,23 +228,21 @@ sniffio==1.3.0
     #   anyio
     #   httpcore
     #   httpx
-starlette==0.26.1
-    # via fastapi
-sympy==1.11.1
+sympy==1.12
     # via
     #   onnxruntime
     #   torch
-timm==0.6.13
+timm==0.9.2
     # via effdet
 tokenizers==0.13.3
     # via transformers
-torch==2.0.0
+torch==2.0.1
     # via
     #   effdet
     #   layoutparser
     #   timm
     #   torchvision
-torchvision==0.15.1
+torchvision==0.15.2
     # via
     #   effdet
     #   layoutparser
@@ -262,22 +254,22 @@ tqdm==4.65.0
     #   iopath
     #   nltk
     #   transformers
-transformers==4.28.1
+transformers==4.29.2
     # via unstructured-inference
+typer==0.9.0
+    # via argilla
 typing-extensions==4.5.0
     # via
     #   huggingface-hub
     #   iopath
     #   pydantic
     #   rich
-    #   starlette
     #   torch
-unstructured-inference==0.4.4
+    #   typer
+unstructured-inference==0.5.0
     # via unstructured (setup.py)
 urllib3==2.0.2
     # via requests
-uvicorn==0.22.0
-    # via unstructured-inference
 wand==0.6.11
     # via pdfplumber
 wrapt==1.14.1

setup.py

@@ -39,6 +39,7 @@
         "Programming Language :: Python :: 3",
         "Programming Language :: Python :: 3.8",
         "Programming Language :: Python :: 3.9",
+        "Programming Language :: Python :: 3.10",
         "Topic :: Scientific/Engineering :: Artificial Intelligence",
     ],
     author="Unstructured Technologies",
@@ -63,6 +64,7 @@
         "python-pptx",
         "python-magic",
         "markdown",
+        "tabulate",
         "requests",
         # NOTE(robinson) - The following dependencies are pinned
         # to address security scans
@@ -77,7 +79,7 @@
             "transformers",
         ],
         "local-inference": [
-            "unstructured-inference==0.4.4",
+            "unstructured-inference==0.5.0",
         ],
         "s3": ["s3fs", "fsspec"],
         "azure": ["adlfs", "fsspec"],

test_unstructured/partition/test_auto.py

@@ -291,12 +291,12 @@ def test_auto_partition_pdf_from_filename(pass_file_filename, content_type):
     assert isinstance(elements[0], Title)
     assert elements[0].text.startswith("LayoutParser")
 
-    assert isinstance(elements[1], NarrativeText)
-    assert elements[1].text.startswith("Zejiang Shen")
-
     assert elements[0].metadata.filename == os.path.basename(filename)
     assert elements[0].metadata.file_directory == os.path.split(filename)[0]
 
+    assert isinstance(elements[1], NarrativeText)
+    assert elements[1].text.startswith("Shen")
+
 
 def test_auto_partition_pdf_uses_table_extraction():
     filename = os.path.join(EXAMPLE_DOCS_DIRECTORY, "layout-parser-paper-fast.pdf")
@@ -346,7 +346,7 @@ def test_auto_partition_pdf_from_file(pass_file_filename, content_type):
     assert elements[0].text.startswith("LayoutParser")
 
     assert isinstance(elements[1], NarrativeText)
-    assert elements[1].text.startswith("Zejiang Shen")
+    assert elements[1].text.startswith("Shen")
 
 
 def test_partition_pdf_doesnt_raise_warning():