[build] Fix the nightly builds

- Replace installation of `pytorch-cuda` with `cuda`, as the
former package has been recently slimmed down to remove `nvcc`
and other CUDA dependencies required for FBGEMM build (see
pytorch/pytorch#91334)

- Add `pytorch` package installation checks to make sure we are
installing the GPU variant of the package

- Add `cuda` package installation checks to make sure we can
access `nvcc` and `cuda_runtime.h`

- Remove Python 3.7 builds from the nightlies

.github/scripts/build_wheel.bash

@@ -18,7 +18,7 @@ usage () {
   echo "-v                  : verbose"
   echo "-h                  : help"
   echo "PACKAGE_NAME        : output package name (e.g., fbgemm_gpu_nightly)"
-  echo "PYTHON_VERSION      : Python version (e.g., 3.7, 3.8, 3.10)"
+  echo "PYTHON_VERSION      : Python version (e.g., 3.8, 3.9, 3.10)"
   echo "PYTORCH_CHANNEL_NAME: PyTorch's channel name (e.g., pytorch-nightly, pytorch-test (=pre-release), pytorch (=stable release))"
   echo "PYTORCH_CUDA_VERSION: PyTorch's CUDA version (e.g., 11.6, 11.7)"
   echo "MINICONDA_PREFIX    : path to install Miniconda (default: \$HOME/miniconda)"

.github/scripts/test_torchrec.bash

@@ -20,7 +20,7 @@ usage () {
   echo "PACKAGE_NAME        : output package name of TorchRec (e.g., torchrec_nightly)"
   echo "                      Note: TorchRec is sensitive to its package name"
   echo "                      e.g., torchrec needs fbgemm-gpu while torchrec_nightly needs fbgemm-gpu-nightly"
-  echo "PYTHON_VERSION      : Python version (e.g., 3.7, 3.8, 3.10)"
+  echo "PYTHON_VERSION      : Python version (e.g., 3.8, 3.9, 3.10)"
   echo "PYTORCH_CHANNEL_NAME: PyTorch's channel name (e.g., pytorch-nightly, pytorch-test (=pre-release), pytorch (=stable release))"
   echo "PYTORCH_CUDA_VERSION: PyTorch's CUDA version (e.g., 11.6, 11.7)"
   echo "FBGEMM_WHEEL_PATH   : path to FBGEMM_GPU's wheel file"

.github/scripts/test_wheel.bash

@@ -17,7 +17,7 @@ usage () {
   echo "Usage: bash test_wheel.bash -p PYTHON_VERSION -P PYTORCH_CHANNEL_NAME -c PYTORCH_CUDA_VERSION -w FBGEMM_WHEEL_PATH [-m MINICONDA_PREFIX] [-v] [-h]"
   echo "-v                  : verbose"
   echo "-h                  : help"
-  echo "PYTHON_VERSION      : Python version (e.g., 3.7, 3.8, 3.10)"
+  echo "PYTHON_VERSION      : Python version (e.g., 3.8, 3.9, 3.10)"
   echo "PYTORCH_CHANNEL_NAME: PyTorch's channel name (e.g., pytorch-nightly, pytorch-test (=pre-release), pytorch (=stable release))"
   echo "PYTORCH_CUDA_VERSION: PyTorch's CUDA version (e.g., 11.6, 11.7)"
   echo "FBGEMM_WHEEL_PATH   : path to FBGEMM_GPU's wheel file"

.github/workflows/fbgemm_nightly_build.yml

@@ -19,10 +19,6 @@ jobs:
     strategy:
       matrix:
         include:
-         - os: linux.2xlarge
-           python-version: "3.7"
-           python-tag: "py37"
-           cuda-tag: "cu11"
          - os: linux.2xlarge
            python-version: "3.8"
            python-tag: "py38"
@@ -70,7 +66,16 @@ jobs:
     - name: Install PyTorch and CUDA
       shell: bash
       run: |
-        conda install -n build_binary -y pytorch pytorch-cuda=11.7 -c pytorch-nightly -c nvidia
+        conda install -n build_binary -y pytorch cuda -c pytorch-nightly -c "nvidia/label/cuda-11.7.1"
+
+        # Ensure that CUDA is properly installed
+        conda run -n build_binary which nvcc
+        conda run -n build_binary test -n "$(find $HOME/miniconda -name cuda_runtime.h)"
+        echo "Successfully installed CUDA"
+
+        # Ensure that the PyTorch nightly build is the GPU variant (i.e. contains cuDNN)
+        conda list -n build_binary pytorch | grep cudnn
+        echo "Successfully installed pytorch-nightly"
     - name: Install Other Dependencies
       shell: bash
       run: |
@@ -130,7 +135,7 @@ jobs:
     strategy:
       matrix:
         os: [linux.g5.4xlarge.nvidia.gpu]
-        python-version: ["3.7", "3.8", "3.9", "3.10"]
+        python-version: ["3.8", "3.9", "3.10"]
         cuda-tag: ["cu11"]
     needs: build_on_cpu
     steps:
@@ -217,7 +222,16 @@ jobs:
     - name: Install PyTorch using Conda
       shell: bash
       run: |
-        conda install -n build_binary -y pytorch pytorch-cuda=11.7 -c pytorch-nightly -c nvidia
+        conda install -n build_binary -y pytorch cuda -c pytorch-nightly -c "nvidia/label/cuda-11.7.1"
+
+        # Ensure that CUDA is properly installed
+        conda run -n build_binary which nvcc
+        conda run -n build_binary test -n "$(find $HOME/miniconda -name cuda_runtime.h)"
+        echo "Successfully installed CUDA"
+
+        # Ensure that the PyTorch nightly build is the GPU variant (i.e. contains cuDNN)
+        conda list -n build_binary pytorch | grep cudnn
+        echo "Successfully installed pytorch-nightly"
     # download wheel from GHA
     - name: Download wheel
       uses: actions/download-artifact@v2