Merge branch 'main' into main

test/test_artifacts/v0/gpu-dependencies.test.Dockerfile

@@ -0,0 +1,11 @@
+ARG SAGEMAKER_DISTRIBUTION_IMAGE
+FROM $SAGEMAKER_DISTRIBUTION_IMAGE
+
+ARG MAMBA_DOCKERFILE_ACTIVATE=1
+
+# Execute cuda valudaiton script:
+# 1. Check if TensorFlow is installed with CUDA support for GPU image
+# 2. Check if Pytorch is installed with CUDA support for GPU image
+COPY --chown=$MAMBA_USER:$MAMBA_USER scripts/cuda_validation.py .
+RUN chmod +x cuda_validation.py
+RUN python3 cuda_validation.py

test/test_artifacts/v0/scripts/cuda_validation.py

@@ -0,0 +1,46 @@
+# Verify Tensorflow CUDA
+import tensorflow as tf
+
+cuda_available = tf.test.is_built_with_cuda()
+if not cuda_available:
+    raise Exception("TensorFlow is installed without CUDA support for GPU image build.")
+print("TensorFlow is built with CUDA support.")
+
+
+# Verify Pytorch is installed with CUDA version
+import subprocess
+
+# Run the micromamba list command and capture the output
+result = subprocess.run(["micromamba", "list"], stdout=subprocess.PIPE, text=True)
+
+# Split the output into lines
+package_lines = result.stdout.strip().split("\n")
+
+# Find the PyTorch entry
+pytorch_entry = None
+for line in package_lines:
+    dependency_info = line.strip().split()
+    if dependency_info and dependency_info[0] == "pytorch":
+        pytorch_entry = line.split()
+        break
+
+# If PyTorch is installed, print its information
+if pytorch_entry:
+    package_name = pytorch_entry[0]
+    package_version = pytorch_entry[1]
+    package_build = pytorch_entry[2]
+    print(f"PyTorch: {package_name} {package_version} {package_build}")
+# Raise exception if CUDA is not detected
+if "cuda" not in package_build:
+    raise Exception("Pytorch is installed without CUDA support for GPU image build.")
+
+# Verify Pytorch has CUDA working properly
+# Because this function only works on a GPU instance, so it may fail in local test
+# To test manually on a GPU instance, run: "docker run --gpus all <image id>"
+import torch
+
+if not torch.cuda.is_available():
+    raise Exception(
+        "Pytorch is installed with CUDA support but not working in current environment. \
+                    Make sure to execute this test case in GPU environment if you are not"
+    )

test/test_artifacts/v1/gpu-dependencies.test.Dockerfile

@@ -0,0 +1,11 @@
+ARG SAGEMAKER_DISTRIBUTION_IMAGE
+FROM $SAGEMAKER_DISTRIBUTION_IMAGE
+
+ARG MAMBA_DOCKERFILE_ACTIVATE=1
+
+# Execute cuda valudaiton script:
+# 1. Check if TensorFlow is installed with CUDA support for GPU image
+# 2. Check if Pytorch is installed with CUDA support for GPU image
+COPY --chown=$MAMBA_USER:$MAMBA_USER scripts/cuda_validation.py .
+RUN chmod +x cuda_validation.py
+RUN python3 cuda_validation.py

test/test_artifacts/v1/scripts/cuda_validation.py

@@ -0,0 +1,46 @@
+# Verify Tensorflow CUDA
+import tensorflow as tf
+
+cuda_available = tf.test.is_built_with_cuda()
+if not cuda_available:
+    raise Exception("TensorFlow is installed without CUDA support for GPU image build.")
+print("TensorFlow is built with CUDA support.")
+
+
+# Verify Pytorch is installed with CUDA version
+import subprocess
+
+# Run the micromamba list command and capture the output
+result = subprocess.run(["micromamba", "list"], stdout=subprocess.PIPE, text=True)
+
+# Split the output into lines
+package_lines = result.stdout.strip().split("\n")
+
+# Find the PyTorch entry
+pytorch_entry = None
+for line in package_lines:
+    dependency_info = line.strip().split()
+    if dependency_info and dependency_info[0] == "pytorch":
+        pytorch_entry = line.split()
+        break
+
+# If PyTorch is installed, print its information
+if pytorch_entry:
+    package_name = pytorch_entry[0]
+    package_version = pytorch_entry[1]
+    package_build = pytorch_entry[2]
+    print(f"PyTorch: {package_name} {package_version} {package_build}")
+# Raise exception if CUDA is not detected
+if "cuda" not in package_build:
+    raise Exception("Pytorch is installed without CUDA support for GPU image build.")
+
+# Verify Pytorch has CUDA working properly
+# Because this function only works on a GPU instance, so it may fail in local test
+# To test manually on a GPU instance, run: "docker run --gpus all <image id>"
+import torch
+
+if not torch.cuda.is_available():
+    raise Exception(
+        "Pytorch is installed with CUDA support but not working in current environment. \
+                    Make sure to execute this test case in GPU environment if you are not"
+    )

test/test_artifacts/v2/gpu-dependencies.test.Dockerfile

@@ -0,0 +1,11 @@
+ARG SAGEMAKER_DISTRIBUTION_IMAGE
+FROM $SAGEMAKER_DISTRIBUTION_IMAGE
+
+ARG MAMBA_DOCKERFILE_ACTIVATE=1
+
+# Execute cuda valudaiton script:
+# 1. Check if TensorFlow is installed with CUDA support for GPU image
+# 2. Check if Pytorch is installed with CUDA support for GPU image
+COPY --chown=$MAMBA_USER:$MAMBA_USER scripts/cuda_validation.py .
+RUN chmod +x cuda_validation.py
+RUN python3 cuda_validation.py

test/test_artifacts/v2/scripts/cuda_validation.py

@@ -0,0 +1,46 @@
+# Verify Tensorflow CUDA
+import tensorflow as tf
+
+cuda_available = tf.test.is_built_with_cuda()
+if not cuda_available:
+    raise Exception("TensorFlow is installed without CUDA support for GPU image build.")
+print("TensorFlow is built with CUDA support.")
+
+
+# Verify Pytorch is installed with CUDA version
+import subprocess
+
+# Run the micromamba list command and capture the output
+result = subprocess.run(["micromamba", "list"], stdout=subprocess.PIPE, text=True)
+
+# Split the output into lines
+package_lines = result.stdout.strip().split("\n")
+
+# Find the PyTorch entry
+pytorch_entry = None
+for line in package_lines:
+    dependency_info = line.strip().split()
+    if dependency_info and dependency_info[0] == "pytorch":
+        pytorch_entry = line.split()
+        break
+
+# If PyTorch is installed, print its information
+if pytorch_entry:
+    package_name = pytorch_entry[0]
+    package_version = pytorch_entry[1]
+    package_build = pytorch_entry[2]
+    print(f"PyTorch: {package_name} {package_version} {package_build}")
+# Raise exception if CUDA is not detected
+if "cuda" not in package_build:
+    raise Exception("Pytorch is installed without CUDA support for GPU image build.")
+
+# Verify Pytorch has CUDA working properly
+# Because this function only works on a GPU instance, so it may fail in local test
+# To test manually on a GPU instance, run: "docker run --gpus all <image id>"
+import torch
+
+if not torch.cuda.is_available():
+    raise Exception(
+        "Pytorch is installed with CUDA support but not working in current environment. \
+                    Make sure to execute this test case in GPU environment if you are not"
+    )

test/test_dockerfile_based_harness.py

@@ -88,6 +88,7 @@ def test_dockerfiles_for_cpu(
         ("mlflow.test.Dockerfile", ["mlflow"]),
         ("sagemaker-mlflow.test.Dockerfile", ["sagemaker-mlflow"]),
         ("jupyter-activity-monitor-extension.test.Dockerfile", ["jupyter-activity-monitor-extension"]),
+        ("gpu-dependencies.test.Dockerfile", ["pytorch", "tensorflow"]),
     ],
 )
 def test_dockerfiles_for_gpu(