Merge branch 'master' into master

docker/Dockerfile

@@ -47,6 +47,7 @@ RUN --mount=type=cache,id=apt-dev,target=/var/cache/apt \
         python$PYTHON_VERSION-venv \
         openjdk-17-jdk \
         curl \
+        git \
     && rm -rf /var/lib/apt/lists/*
 
 # Make the virtual environment and "activating" it by adding it first to the path.
@@ -62,26 +63,29 @@ RUN export USE_CUDA=1
 
 ARG CUDA_VERSION=""
 
-RUN TORCH_VER=$(curl --silent --location https://pypi.org/pypi/torch/json | python -c "import sys, json, pkg_resources; releases = json.load(sys.stdin)['releases']; print(sorted(releases, key=pkg_resources.parse_version)[-1])") && \
-    TORCH_VISION_VER=$(curl --silent --location https://pypi.org/pypi/torchvision/json | python -c "import sys, json, pkg_resources; releases = json.load(sys.stdin)['releases']; print(sorted(releases, key=pkg_resources.parse_version)[-1])") && \
+RUN git clone --depth 1 https://github.com/pytorch/serve.git
+
+WORKDIR "serve"
+
+RUN \
     if echo "$BASE_IMAGE" | grep -q "cuda:"; then \
         # Install CUDA version specific binary when CUDA version is specified as a build arg
         if [ "$CUDA_VERSION" ]; then \
-            python -m pip install --no-cache-dir torch==$TORCH_VER+$CUDA_VERSION torchvision==$TORCH_VISION_VER+$CUDA_VERSION -f https://download.pytorch.org/whl/torch_stable.html; \
-        # Install the binary with the latest CUDA version support
+            python ./ts_scripts/install_dependencies.py --cuda $CUDA_VERSION; \
+        # Install the binary with the latest CPU image on a CUDA base image
         else \
-            python -m pip install --no-cache-dir torch torchvision; \
+            python ./ts_scripts/install_dependencies.py; \
         fi; \
-        python -m pip install --no-cache-dir -r https://raw.githubusercontent.com/pytorch/serve/master/requirements/common.txt; \
     # Install the CPU binary
     else \
-        python -m pip install --no-cache-dir torch==$TORCH_VER+cpu torchvision==$TORCH_VISION_VER+cpu -f https://download.pytorch.org/whl/torch_stable.html; \
+        python ./ts_scripts/install_dependencies.py; \
     fi
 
-RUN python -m pip install --no-cache-dir captum torchtext torchserve torch-model-archiver pyyaml
+# Make sure latest version of torchserve is uploaded before running this
+RUN python -m pip install --no-cache-dir torchserve torch-model-archiver torch-workflow-archiver
 
 # Final image for production
-FROM ${BASE_IMAGE} AS runtime-image 
+FROM ${BASE_IMAGE} AS runtime-image
 # Re-state ARG PYTHON_VERSION to make it active in this build-stage (uses default define at the top)
 ARG PYTHON_VERSION
 ENV PYTHONUNBUFFERED TRUE

examples/Huggingface_Transformers/README.md

@@ -10,7 +10,7 @@ We borrowed ideas to write a custom handler for transformers from tutorial prese
 
 To get started [install Torchserve](https://github.com/pytorch/serve) and then
 
- `pip install transformers==4.6.0`
+ `pip install -r requirements.txt`
 
 ### Objectives
 1. How to package a transformer into a torch model archive (.mar) file (eager mode or Torchscript) with `torch-model-archiver`
@@ -315,7 +315,6 @@ When a json file is passed as a request format to the curl, Torchserve unwraps t
 
 In the setup_config.json, specify `"BetterTransformer":true,`.
 
-Note: make sure to install [HuggingFace Optimum] `pip install optimum`
 
 [Better Transformer](https://pytorch.org/blog/a-better-transformer-for-fast-transformer-encoder-inference/) from PyTorch is integrated into [Huggingface Optimum](https://huggingface.co/docs/optimum/bettertransformer/overview) that bring major speedups for many of encoder models on different modalities (text, image, audio). It is a one liner API that we have also added in the `Transformer_handler_generalized.py` in this example as well. That as shown above you just need to set `"BetterTransformer":true,` in the setup_config.json.
 

examples/Huggingface_Transformers/Transformer_handler_generalized.py

@@ -83,14 +83,10 @@ def initialize(self, ctx):
                 logger.warning("Missing the operation mode.")
             # Using the Better Transformer integration to speedup the inference
             if self.setup_config["BetterTransformer"]:
-                try:
-                    from optimum.bettertransformer import BetterTransformer
+                from optimum.bettertransformer import BetterTransformer
 
+                try:
                     self.model = BetterTransformer.transform(self.model)
-                except ImportError as error:
-                    logger.warning(
-                        "HuggingFace Optimum is not installed. Proceeding without BetterTransformer"
-                    )
                 except RuntimeError as error:
                     logger.warning(
                         "HuggingFace Optimum is not supporting this model,for the list of supported models, please refer to this doc,https://huggingface.co/docs/optimum/bettertransformer/overview"
@@ -377,7 +373,6 @@ def get_insights(self, input_batch, text, target):
             self.setup_config["mode"] == "sequence_classification"
             or self.setup_config["mode"] == "token_classification"
         ):
-
             attributions, delta = self.lig.attribute(
                 inputs=input_ids,
                 baselines=ref_input_ids,

examples/Huggingface_Transformers/requirements.txt

@@ -0,0 +1,2 @@
+transformers
+optimum

model-archiver/model_archiver/version.txt

@@ -1 +1 @@
-0.7.1
+0.8.0

ts/version.txt

@@ -1 +1 @@
-0.7.1
+0.8.0

workflow-archiver/workflow_archiver/version.txt

@@ -1 +1 @@
-0.2.7
+0.2.8