Don't install torch by default (#472)

* Don't use torch by default

* Ensure torch install on CI

* Run OCR benchmarks on relevant pull rquests too

.github/workflows/ocr-benchmarks.yml

@@ -1,5 +1,9 @@
 name: OCR Poetry Benchmark Tests
 on:
+  pull_request:
+    paths:
+      - .github/workflows/ocr-benchmarks.yml
+      - OCR/**
   push:
     branches:
       - main
@@ -16,31 +20,13 @@ jobs:
           python-version: "3.10"
       - name: Install dependencies
         run: |
+          sudo apt update && sudo apt install tesseract-ocr-eng tesseract-ocr
           cd OCR/tests
           python -m pip install --upgrade pip
           pip install poetry
           pip install pytest pytest-benchmark
-          poetry install --with dev
+          poetry install --with dev --with torch
       - name: Run tests
         run: |
           cd OCR/tests
           poetry run pytest benchmark_test.py -v --benchmark-json output.json
-      # - name: Store benchmark result
-      #   uses: benchmark-action/github-action-benchmark@v1
-      #   with:
-      #     tool: 'pytest'
-      #     output-file-path: OCR/tests/output.json
-      #     auto-push: false
-      # - name: Push benchmark result
-      #   run: git push 'https://CDCgov:${{ secrets.GITHUB_TOKEN }}@github.com/CDCgov/IDWA.git' gh-pages:gh-pages
-      # - name: Fetch data.js from gh-pages
-      #   run: |
-      #     git config --global user.name 'GitHub Action'
-      #     git config --global user.email '[email protected]'
-      #     git fetch
-      #     git checkout gh-pages
-      #     rm ./dev/bench/index.html
-      #     cp ./dev/bench/index-template.html ./dev/bench/index.html
-      #     git add .
-      #     git commit -m "template to index"
-      #     git push origin gh-pages

.github/workflows/ocr-tests.yml

@@ -20,6 +20,6 @@ jobs:
           sudo apt install tesseract-ocr-eng tesseract-ocr -y
           python -m pip install --upgrade pip
           pip install poetry
-          poetry install --with dev
+          poetry install --with dev --with torch
       - name: Run tests
         run: poetry run pytest tests/ocr_test.py tests/segmentation_template_test.py tests/alignment_test.py

OCR/dev-dockerfile

@@ -12,4 +12,4 @@ COPY ./pyproject.toml /ocr/pyproject.toml
 COPY ./poetry.lock /ocr/poetry.lock
 
 WORKDIR /ocr
-RUN poetry install && poetry cache list | xargs -n1 poetry cache clear --all
+RUN poetry install --without=torch && poetry cache list | xargs -n1 poetry cache clear --all

OCR/ocr/services/alignment/backends/random_perspective_transform.py

@@ -4,7 +4,6 @@
 
 from pathlib import Path
 
-import torch
 import numpy as np
 import cv2 as cv
 from PIL import Image
@@ -20,6 +19,8 @@ def make_transform(self, distortion_scale: float) -> object:
         """
         Create a transformation matrix for a random perspective transform.
         """
+        import torch
+
         # From torchvision. BSD 3-clause
         height = self.image.height
         width = self.image.width

OCR/pyproject.toml

@@ -11,11 +11,13 @@ numpy = "^1.26.4"
 opencv-python = "^4.9.0.80"
 levenshtein = "^0.25.1"
 fastapi = {extras = ["standard"], version = "^0.112.1"}
-transformers = {extras = ["torch"], version = "^4.45.1"}
 pillow = "^10.3.0"
 datasets = "^3.0.1"
 tesserocr = "^2.7.1"
 
+[tool.poetry.group.torch.dependencies]
+transformers = {extras = ["torch"], version = "^4.45.1"}
+
 [tool.poetry.group.dev.dependencies]
 lxml = "^5.3.0"
 docopt = "^0.6.2"

dev-env.yaml

@@ -8,7 +8,7 @@ services:
     volumes:
       - ./OCR:/ocr
     restart: "no"
-    command: "/bin/sh -c 'poetry install && poetry run fastapi dev ocr/api.py --host 0.0.0.0 --port 8000'" 
+    command: "/bin/sh -c 'poetry install --without=torch --sync && poetry run fastapi dev ocr/api.py --host 0.0.0.0 --port 8000'"
   frontend:
     build:
       context: ./frontend