Dockerfile refacto: split requirements and source code layers

To optimize build time and enhance layer reuse

Signed-off-by: Raphael Glon <[email protected]>

Author: oOraph

dockerfiles/pytorch/Dockerfile

@@ -32,8 +32,7 @@ RUN apt-get update && \
     && apt-get clean autoremove --yes \
     && rm -rf /var/lib/{apt,dpkg,cache,log}
 
-# Copying only necessary files as filtered by .dockerignore
-COPY . .
+RUN mkdir -p /var/lib/dpkg && touch /var/lib/dpkg/status
 
 # Set Python 3.11 as the default python version
 RUN update-alternatives --install /usr/bin/python3 python3 /usr/bin/python3.11 1 && \
@@ -47,13 +46,20 @@ RUN curl -O https://bootstrap.pypa.io/get-pip.py && \
 # Upgrade pip
 RUN pip install --no-cache-dir --upgrade pip
 
+COPY requirements.txt .
+RUN pip install -r requirements.txt
+
+# Copying only necessary files as filtered by .dockerignore
+COPY . .
 # Install wheel and setuptools
 RUN pip install --no-cache-dir --upgrade pip ".[torch,st,diffusers]"
 
 # Copy application
 COPY src/huggingface_inference_toolkit huggingface_inference_toolkit
 COPY src/huggingface_inference_toolkit/webservice_starlette.py webservice_starlette.py
 
+RUN rm -rf src/
+
 # Copy entrypoint and change permissions
 COPY --chmod=0755 scripts/entrypoint.sh entrypoint.sh
 

setup.py

@@ -1,7 +1,19 @@
 from __future__ import absolute_import
-
+import os
 from setuptools import find_packages, setup
 
+lib_folder = os.path.dirname(os.path.realpath(__file__))
+requirements_path = f"{lib_folder}/requirements.txt"
+install_requires = [] # Here we'll add: ["gunicorn", "docutils>=0.3", "lxml==0.5a7"]
+if os.path.isfile(requirements_path):
+    with open(requirements_path) as f:
+        install_requires = f.read().splitlines()
+
+test_requirements_path =  f"{lib_folder}/test-requirements.txt"
+if os.path.isfile(test_requirements_path):
+    with open(test_requirements_path) as f:
+        test_requirements = f.read().splitlines()
+
 # We don't declare our dependency on transformers here because we build with
 # different packages for different variants
 
@@ -12,49 +24,14 @@
 # ffmpeg: ffmpeg is required for audio processing. On Ubuntu it can be installed as follows: apt install ffmpeg
 # libavcodec-extra : libavcodec-extra  includes additional codecs for ffmpeg
 
-install_requires = [
-    # Due to an error affecting kenlm and cmake (see https://github.com/kpu/kenlm/pull/464)
-    # Also see the transformers patch for it https://github.com/huggingface/transformers/pull/37091
-    "kenlm@git+https://github.com/kpu/kenlm@ba83eafdce6553addd885ed3da461bb0d60f8df7",
-    "transformers[sklearn,sentencepiece,audio,vision]==4.51.3",
-    "huggingface_hub[hf_transfer,hf_xet]==0.31.1",
-    # vision
-    "Pillow",
-    "librosa",
-    # speech + torchaudio
-    "pyctcdecode>=0.3.0",
-    "phonemizer",
-    "ffmpeg",
-    # web api
-    "starlette",
-    "uvicorn",
-    "gunicorn",
-    "pandas",
-    "orjson",
-    "einops",
-    "timm",
-]
-
 extras = {}
-
 extras["st"] = ["sentence_transformers==4.0.2"]
 extras["diffusers"] = ["diffusers==0.33.1", "accelerate==1.6.0"]
 # Includes `peft` as PEFT requires `torch` so having `peft` as a core dependency
 # means that `torch` will be installed even if the `torch` extra is not specified.
 extras["torch"] = ["torch==2.5.1", "torchvision", "torchaudio", "peft==0.15.1"]
-extras["test"] = [
-    "pytest==7.2.1",
-    "pytest-xdist",
-    "parameterized",
-    "psutil",
-    "datasets",
-    "pytest-sugar",
-    "mock==2.0.0",
-    "docker",
-    "requests",
-    "tenacity",
-]
 extras["quality"] = ["isort", "ruff"]
+extras["test"] = test_requirements
 extras["inf2"] = ["optimum-neuron"]
 extras["google"] = ["google-cloud-storage", "crcmod==1.7"]
 

test-requirements.txt

@@ -0,0 +1,12 @@
+isort
+ruff
+datasets
+docker
+mock==2.0.0
+parameterized
+psutil
+pytest-sugar
+pytest-xdist
+pytest==7.2.1
+requests
+tenacity