Fixes

Author: PastelStorm

.github/workflows/ci.yml

@@ -54,7 +54,7 @@ jobs:
         sudo apt-get update && sudo apt-get install --yes poppler-utils libreoffice
         uv sync --group test --locked
         make install-pandoc
-        make install-nltk-models
+        make install-nlp-models
         sudo add-apt-repository -y ppa:alex-p/tesseract-ocr5
         sudo apt-get install -y tesseract-ocr tesseract-ocr-kor
         tesseract --version

CHANGELOG.md

@@ -1,5 +1,6 @@
 ## 0.1.2
-* Bump all packages (refresh uv.lock)
+* Bump all packages (refresh uv.lock), pulling `unstructured==0.22.12` which replaces NLTK with spaCy
+* Replace `download_nltk_packages` calls with spaCy model pre-download in Makefile, Dockerfile, and CI
 * Switch `uv sync --frozen` to `uv sync --locked` across Dockerfile, Makefile, and CI workflows
 
 ## 0.1.1

Dockerfile

@@ -71,7 +71,7 @@ RUN ARCH=$(uname -m) && \
     cp /tmp/pandoc-${PANDOC_VERSION}/bin/pandoc /home/${USER}/.local/bin/ && \
     rm -rf /tmp/pandoc*
 
-RUN ${PYTHON} -c "from unstructured.nlp.tokenize import download_nltk_packages; download_nltk_packages()" && \
+RUN ${PYTHON} -c "from unstructured.nlp.tokenize import _load_spacy_model; _load_spacy_model()" && \
     ${PYTHON} -c "from unstructured.partition.model_init import initialize; initialize()" && \
     ${PYTHON} -c "from unstructured_inference.models.tables import UnstructuredTableTransformerModel; model = UnstructuredTableTransformerModel(); model.initialize('microsoft/table-transformer-structure-recognition')"
 

Makefile

@@ -13,7 +13,7 @@ help: Makefile
 
 ## install-base:                installs minimum requirements to run the API
 .PHONY: install-base
-install-base: install-base-packages install-nltk-models
+install-base: install-base-packages install-nlp-models
 
 ## install:                     installs all test and dev requirements
 .PHONY: install
@@ -27,9 +27,9 @@ install-base-packages:
 install-test:
 	uv sync --group test --locked
 
-.PHONY: install-nltk-models
-install-nltk-models:
-	uv run python -c "from unstructured.nlp.tokenize import download_nltk_packages; download_nltk_packages()"
+.PHONY: install-nlp-models
+install-nlp-models:
+	uv run python -c "from unstructured.nlp.tokenize import _load_spacy_model; _load_spacy_model()"
 
 ## lock:                        regenerates uv.lock
 .PHONY: lock

prepline_general/api/general.py

@@ -145,7 +145,7 @@ def partition_file_via_api(
     if not request_url:
         raise HTTPException(status_code=500, detail="Parallel mode enabled but no url set!")
 
-    api_key = request.headers.get("unstructured-api-key", default="")
+    api_key = request.headers.get("unstructured-api-key", "")
     partition_kwargs["starting_page_number"] = (
         partition_kwargs.get("starting_page_number", 1) + page_offset
     )

pyproject.toml

@@ -6,6 +6,7 @@ requires-python = ">=3.12"
 dependencies = [
     "unstructured[all-docs] >=0.18.31, <1.0.0",
     "fastapi >=0.128.4, <1.0.0",
+    "python-multipart >=0.0.18",
     "uvicorn >=0.40.0, <1.0.0",
     "backoff >=2.2.1, <3.0.0",
     "pandas >=3.0.0, <4.0.0",