Reduce ocrd/core-cuda

Dockerfile

@@ -6,7 +6,7 @@ ENV DEBIAN_FRONTEND noninteractive
 ENV PYTHONIOENCODING utf8
 ENV LC_ALL=C.UTF-8
 ENV LANG=C.UTF-8
-ENV PIP=pip3
+ENV PIP=pip
 
 WORKDIR /build-ocrd
 COPY ocrd ./ocrd
@@ -24,7 +24,6 @@ RUN apt-get update && apt-get -y install software-properties-common \
     && apt-get update && apt-get -y install \
         ca-certificates \
         python3-dev \
-        python3-pip \
         python3-venv \
         gcc \
         make \
@@ -34,11 +33,11 @@ RUN apt-get update && apt-get -y install software-properties-common \
         sudo \
         git \
     && make deps-ubuntu \
-    && pip3 install --upgrade pip setuptools \
+    && python3 -m venv /usr/local \
+    && hash -r \
+    && pip install --upgrade pip setuptools \
     && make install \
-    && apt-get remove -y gcc \
-    && apt-get autoremove -y \
-    && $FIXUP \
+    && eval $FIXUP \
     && rm -rf /build-ocrd
 
 WORKDIR /data

Dockerfile.cuda

@@ -0,0 +1,61 @@
+ARG BASE_IMAGE
+FROM $BASE_IMAGE
+
+ENV MAMBA_EXE=/usr/local/bin/conda
+ENV MAMBA_ROOT_PREFIX=/conda
+ENV PATH=$MAMBA_ROOT_PREFIX/bin:$PATH
+ENV CONDA_PREFIX=$MAMBA_ROOT_PREFIX
+ENV CONDA_SHLVL='1'
+
+WORKDIR $MAMBA_ROOT_PREFIX
+
+RUN curl -Ls https://micro.mamba.pm/api/micromamba/linux-64/latest | tar -xvj bin/micromamba
+RUN mv bin/micromamba $MAMBA_EXE
+RUN hash -r
+RUN mkdir -p $CONDA_PREFIX/lib $CONDA_PREFIX/include
+RUN echo $CONDA_PREFIX/lib >> /etc/ld.so.conf.d/conda.conf
+# Get CUDA toolkit, including compiler and libraries with dev.
+# The nvidia channels do not provide (recent) cudnn (needed for Torch, TF etc):
+#RUN conda install -c nvidia/label/cuda-11.8.0 cuda && conda clean -a
+# The conda-forge channel has cudnn but no cudatoolkit-dev anymore,
+# so let's combine nvidia and conda-forge (will be same lib versions, no waste of space):
+RUN conda install -c nvidia/label/cuda-11.8.0 \
+                  cuda-nvcc \
+                  cuda-cccl \
+ && conda clean -a \
+ && find $CONDA_PREFIX -name "*_static.a" -delete
+                  # cuda-cudart-dev \
+                  # cuda-libraries-dev \
+#RUN conda install -c conda-forge \
+#          cudatoolkit=11.8.0 \
+#          cudnn=8.8.* && \
+#    conda clean -a && \
+#    find $CONDA_PREFIX -name "*_static.a" -delete
+# Since Torch will pull in the CUDA libraries (as Python pkgs) anyway,
+# let's jump the shark and pull these via NGC index directly,
+# but then share them with the rest of the system so native compilation/linking
+# works, too:
+RUN pip3 install nvidia-pyindex \
+ && pip3 install nvidia-cudnn-cu11==8.6.0.163 \
+                 nvidia-cublas-cu11 \
+                 nvidia-cusparse-cu11 \
+                 nvidia-cusolver-cu11 \
+                 nvidia-curand-cu11 \
+                 nvidia-cufft-cu11 \
+                 nvidia-cuda-runtime-cu11 \
+                 nvidia-cuda-nvrtc-cu11 \
+ && for pkg in cudnn cublas cusparse cusolver curand cufft cuda_runtime cuda_nvrtc; do \
+        for lib in /usr/local/lib/python3.8/site-packages/nvidia/$pkg/lib/lib*.so.*; do \
+            base=$(basename $lib); \
+            ln -s $lib $CONDA_PREFIX/lib/$base.so; \
+            ln -s $lib $CONDA_PREFIX/lib/${base%.so.*}.so; \
+        done \
+     && ln -s /usr/local/lib/python3.8/site-packages/nvidia/$pkg/include/* $CONDA_PREFIX/include/; \
+    done \
+ && ldconfig
+# gputil/nvidia-smi would be nice, too – but that drags in Python as a conda dependency...
+
+WORKDIR /data
+
+CMD ["/usr/local/bin/ocrd", "--help"]
+

Makefile

@@ -32,28 +32,18 @@ help:
 	@echo "    docs-clean     Clean docs"
 	@echo "    docs-coverage  Calculate docstring coverage"
 	@echo "    docker         Build docker image"
-	@echo "    docker-cuda    Build docker GPU / CUDA image"
-	@echo "    cuda-ubuntu    Install native CUDA toolkit in different versions"
+	@echo "    docker-cuda    Build docker image for GPU / CUDA"
 	@echo "    pypi           Build wheels and source dist and twine upload them"
 	@echo ""
 	@echo "  Variables"
 	@echo ""
-	@echo "    DOCKER_TAG         Docker tag. Default: '$(DOCKER_TAG)'."
-	@echo "    DOCKER_BASE_IMAGE  Docker base image. Default: '$(DOCKER_BASE_IMAGE)'."
+	@echo "    DOCKER_TAG         Docker target image tag. Default: '$(DOCKER_TAG)'."
+	@echo "    DOCKER_BASE_IMAGE  Docker source image tag. Default: '$(DOCKER_BASE_IMAGE)'."
 	@echo "    DOCKER_ARGS        Additional arguments to docker build. Default: '$(DOCKER_ARGS)'"
 	@echo "    PIP_INSTALL        pip install command. Default: $(PIP_INSTALL)"
 
 # END-EVAL
 
-# Docker tag. Default: '$(DOCKER_TAG)'.
-DOCKER_TAG = ocrd/core
-
-# Docker base image. Default: '$(DOCKER_BASE_IMAGE)'.
-DOCKER_BASE_IMAGE = ubuntu:20.04
-
-# Additional arguments to docker build. Default: '$(DOCKER_ARGS)'
-DOCKER_ARGS = 
-
 # pip install command. Default: $(PIP_INSTALL)
 PIP_INSTALL = $(PIP) install
 
@@ -68,12 +58,13 @@ deps-test:
 
 # (Re)install the tool
 install:
-	$(PIP) install -U pip wheel setuptools fastentrypoints
+	$(PIP) install -U pip wheel setuptools
 	@# speedup for end-of-life builds
+	@# we cannot use pip config here due to pip#11988
 	if $(PYTHON) -V | fgrep -e 3.5 -e 3.6; then $(PIP) install --prefer-binary opencv-python-headless numpy; fi
 	for mod in $(BUILD_ORDER);do (cd $$mod ; $(PIP_INSTALL) .);done
 	@# workaround for shapely#1598
-	$(PIP) install --no-binary shapely --force-reinstall shapely
+	$(PIP) config set global.no-binary shapely
 
 # Install with pip install -e
 install-dev: uninstall
@@ -214,40 +205,22 @@ pyclean:
 
 .PHONY: docker docker-cuda
 
+# Additional arguments to docker build. Default: '$(DOCKER_ARGS)'
+DOCKER_ARGS = 
+
 # Build docker image
-docker docker-cuda:
-	docker build -t $(DOCKER_TAG) --build-arg BASE_IMAGE=$(DOCKER_BASE_IMAGE) $(DOCKER_ARGS) .
+docker: DOCKER_BASE_IMAGE = ubuntu:20.04
+docker: DOCKER_TAG = ocrd/core
+docker: DOCKER_FILE = Dockerfile
 
-# Build docker GPU / CUDA image
-docker-cuda: DOCKER_BASE_IMAGE = nvidia/cuda:11.3.1-cudnn8-runtime-ubuntu20.04
+docker-cuda: DOCKER_BASE_IMAGE = ocrd/core
 docker-cuda: DOCKER_TAG = ocrd/core-cuda
-docker-cuda: DOCKER_ARGS += --build-arg FIXUP="make cuda-ubuntu cuda-ldconfig"
+docker-cuda: DOCKER_FILE = Dockerfile.cuda
 
-#
-# CUDA
-#
+docker-cuda: docker
 
-.PHONY: cuda-ubuntu cuda-ldconfig
-
-# Install native CUDA toolkit in different versions
-cuda-ubuntu: cuda-ldconfig
-	apt-get -y install --no-install-recommends cuda-runtime-11-0 cuda-runtime-11-1 cuda-runtime-11-3 cuda-runtime-11-7 cuda-runtime-12-1
-
-cuda-ldconfig: /etc/ld.so.conf.d/cuda.conf
-	ldconfig
-
-/etc/ld.so.conf.d/cuda.conf:
-	@echo > $@
-	@echo /usr/local/cuda-11.0/lib64 >> $@
-	@echo /usr/local/cuda-11.0/targets/x86_64-linux/lib >> $@
-	@echo /usr/local/cuda-11.1/lib64 >> $@
-	@echo /usr/local/cuda-11.1/targets/x86_64-linux/lib >> $@
-	@echo /usr/local/cuda-11.3/lib64 >> $@
-	@echo /usr/local/cuda-11.3/targets/x86_64-linux/lib >> $@
-	@echo /usr/local/cuda-11.7/lib64 >> $@
-	@echo /usr/local/cuda-11.7/targets/x86_64-linux/lib >> $@
-	@echo /usr/local/cuda-12.1/lib64 >> $@
-	@echo /usr/local/cuda-12.1/targets/x86_64-linux/lib >> $@
+docker docker-cuda: 
+	docker build --progress=plain -f $(DOCKER_FILE) -t $(DOCKER_TAG) --build-arg BASE_IMAGE=$(DOCKER_BASE_IMAGE) $(DOCKER_ARGS) .
 
 # Build wheels and source dist and twine upload them
 pypi: uninstall install

ocrd/ocrd/constants.py

@@ -1,7 +1,7 @@
 """
 Constants for ocrd.
 """
-from ocrd_utils.package_resources import resource_filename
+from pkg_resources import resource_filename
 
 __all__ = [
     'TMP_PREFIX',

ocrd/ocrd/processor/builtin/dummy_processor.py

@@ -1,6 +1,6 @@
 # pylint: disable=missing-module-docstring,invalid-name
 from os.path import join, basename
-from ocrd_utils.package_resources import resource_string
+from pkg_resources import resource_string
 
 import click
 
@@ -17,7 +17,7 @@
 )
 from ocrd_modelfactory import page_from_file
 
-OCRD_TOOL = parse_json_string_with_comments(resource_string(__name__, 'ocrd-tool.json').decode('utf8'))
+OCRD_TOOL = parse_json_string_with_comments(resource_string(__name__, 'dummy/ocrd-tool.json').decode('utf8'))
 
 class DummyProcessor(Processor):
     """

ocrd/ocrd/resource_list.yml

@@ -59,19 +59,3 @@ ocrd-sbb-binarize:
     type: archive
     path_in_archive: models
     size: 1654623597
-ocrd-sbb-textline-detector:
-  - url: https://qurator-data.de/sbb_textline_detector/models.tar.gz
-    description: default models provided by github.com/qurator-spk
-    name: default
-    type: archive
-    size: 1194551551
-ocrd-kraken-segment:
-  - url: https://github.com/mittagessen/kraken/raw/master/kraken/blla.mlmodel
-    description: Pretrained baseline segmentation model
-    name: blla.mlmodel
-    size: 5046835
-ocrd-kraken-recognize:
-  - url: https://zenodo.org/record/2577813/files/en_best.mlmodel?download=1
-    name: en_best.mlmodel
-    description: This model has been trained on a large corpus of modern printed English text\naugmented with ~10000 lines of historical pages
-    size: 2930723

ocrd/ocrd/workspace_bagger.py

@@ -9,6 +9,7 @@
 import sys
 from bagit import Bag, make_manifests, _load_tag_file, _make_tag_file, _make_tagmanifest_file  # pylint: disable=no-name-in-module
 from distutils.dir_util import copy_tree
+from pkg_resources import get_distribution
 
 from ocrd_utils import (
     pushd_popd,
@@ -22,7 +23,6 @@
 from ocrd_validators.constants import BAGIT_TXT, TMP_BAGIT_PREFIX, OCRD_BAGIT_PROFILE_URL
 from ocrd_modelfactory import page_from_file
 from ocrd_models.ocrd_page import to_xml
-from ocrd_utils.package_resources import get_distribution
 
 from .workspace import Workspace
 

ocrd/setup.py

@@ -1,5 +1,4 @@
 # -*- coding: utf-8 -*-
-import fastentrypoints
 from setuptools import setup, find_packages
 from ocrd_utils import VERSION
 

ocrd_models/ocrd_models/constants.py

@@ -1,7 +1,7 @@
 """
 Constants for ocrd_models.
 """
-from ocrd_utils.package_resources import resource_string
+from pkg_resources import resource_string
 import re
 
 __all__ = [

ocrd_utils/ocrd_utils/constants.py

@@ -1,12 +1,11 @@
 """
 Constants for ocrd_utils.
 """
+from pkg_resources import get_distribution
 from re import compile as regex_compile
 from os import environ
 from os.path import join, expanduser
 
-from ocrd_utils.package_resources import get_distribution
-
 __all__ = [
     'EXT_TO_MIME',
     'LOG_FORMAT',

ocrd_utils/requirements.txt

@@ -3,6 +3,4 @@ Pillow >= 7.2.0
 # tensorflow versions might require different versions
 numpy
 atomicwrites >= 1.3.0
-importlib_metadata;python_version<'3.8'
-importlib_resources;python_version<'3.9'
 frozendict>=2.3.4

ocrd_validators/ocrd_validators/constants.py

@@ -2,7 +2,7 @@
 Constants for ocrd_validators.
 """
 import yaml
-from ocrd_utils.package_resources import resource_string, resource_filename
+from pkg_resources import resource_string, resource_filename
 
 __all__ = [
     'PROCESSING_SERVER_CONFIG_SCHEMA',