Mlflow update 2209

models/mlflow/README.md

@@ -34,6 +34,18 @@ Example:
 mkdir -p /opt/triton_models
 ```
 
+## Download Morpheus reference models
+
+The Morpheus reference models can be found in the [Morpheus](https://github.com/NVIDIA/Morpheus) repo. A script is provided to fetch the models using git-lfs due to size.
+Before running the MLflow plugin container, you can fetch the models and mount them to the local path on your host (for example, `/opt/triton_models`).
+
+```bash
+git clone https://github.com/NVIDIA/Morpheus.git morpheus
+cd morpheus
+scripts/fetch_data.py fetch models
+cp -RL models /opt/triton_models
+```
+
 ## Start Triton Inference Server in EXPLICIT mode
 
 Use the following command to run Triton with our model
@@ -62,7 +74,7 @@ docker run -it -v /opt/triton_models:/triton_models \
 --gpus '"device=0"' \
 --net=host \
 --rm \
--d mlflow-morpheus:latest
+-d mlflow-triton-plugin:latest
 ```
 
 Open Bash shell in container:
@@ -77,17 +89,6 @@ docker exec -it <container_name> bash
 nohup mlflow server --backend-store-uri sqlite:////tmp/mlflow-db.sqlite --default-artifact-root /mlflow/artifacts --host 0.0.0.0 &
 ```
 
-
-## Download Morpheus reference models
-
-The Morpheus reference models can be found in the [Morpheus](https://github.com/NVIDIA/Morpheus) repo.
-
-```bash
-git clone https://github.com/NVIDIA/Morpheus.git morpheus
-cd morpheus
-scripts/fetch_data.py fetch models
-```
-
 ## Publish reference models to MLflow
 
 The `publish_model_to_mlflow` script is used to publish `triton` flavor models to MLflow. A `triton` flavor model is a directory containing the model files following the [model layout](https://github.com/triton-inference-server/server/blob/main/docs/model_repository.md#repository-layout). Below is an example usage:
@@ -102,7 +103,7 @@ python publish_model_to_mlflow.py \
 ## Deployments
 
 The Triton `mlflow-triton-plugin` is installed on this container and can be used to deploy your models from MLflow to Triton Inference Server. The following are examples of how the plugin is used with the `sid-minibert-onnx` model that we published to MLflow above. For more information about the
-`mlflow-triton-plugin`, please see Triton's [documentation](https://github.com/triton-inference-server/server/tree/r21.12/deploy/mlflow-triton-plugin)
+`mlflow-triton-plugin`, please see Triton's [documentation](https://github.com/triton-inference-server/server/tree/r22.09/deploy/mlflow-triton-plugin)
 
 ### Create Deployment
 

models/mlflow/docker/Dockerfile

@@ -14,9 +14,8 @@
 # limitations under the License.
 
 ARG FROM_IMAGE=gpuci/miniforge-cuda
-ARG CUDA_VER=11.4
+ARG CUDA_VER=11.5
 ARG LINUX_VER=ubuntu20.04
-ARG TRITON_VER=r22.02
 FROM ${FROM_IMAGE}:${CUDA_VER}-runtime-${LINUX_VER} AS base
 
 WORKDIR /mlflow
@@ -31,7 +30,7 @@ RUN sed -i 's/conda activate base/conda activate mlflow/g' ~/.bashrc
 SHELL ["/opt/conda/bin/conda", "run", "-n", "mlflow", "/bin/bash", "-c"]
 
 ARG TRITON_DIR=/mlflow/triton-inference-server
-ARG TRITON_VER=r21.12
+ARG TRITON_VER=r22.07
 
 RUN mkdir ${TRITON_DIR} && \
     cd ${TRITON_DIR} && \
@@ -43,6 +42,9 @@ RUN mkdir ${TRITON_DIR} && \
 RUN ln -sf ${TRITON_DIR}/server/deploy/mlflow-triton-plugin/scripts/publish_model_to_mlflow.py /mlflow && \
     mkdir /mlflow/artifacts
 
+# remedy for CVE-2015-20107
+RUN find / -name '*mailcap*.*py*' | xargs rm
+
 ENV MLFLOW_MODEL_REPO=/mlflow/artifacts
 ENV MLFLOW_TRACKING_URI=sqlite:////tmp/mlflow-db.sqlite
 

models/mlflow/docker/conda/mlflow-env.yml

@@ -21,7 +21,8 @@ dependencies:
   - python=3.8
   - pip
   - onnx
-  - openssl>=1.1.1o
+  - openssl>=1.1.1q
+  - psycopg2<3
   - pip:
+    - boto3
     - mlflow
-