Automatically identify cuda from nvidia-smi in install-requirements (p…

…ytorch#606)

* Automatically identify cuda from nvidia-smi in install-requirements

* Update README.md

---------

Co-authored-by: Michael Gschwind <61328285+mikekgfb@users.noreply.github.com>

README.md

@@ -54,11 +54,7 @@ cd torchchat
 # set up a virtual environment
 python3 -m venv .venv
 source .venv/bin/activate
-```
-
-After the virtual environment is activated, dependencies can be installed.
 
-```bash
 # install dependencies
 ./install_requirements.sh
 ```

install_requirements.sh

@@ -42,10 +42,14 @@ $PIP_EXECUTABLE install -r requirements.txt --extra-index-url https://download.p
 NIGHTLY_VERSION=dev20240422
 
 # The pip repository that hosts nightly torch packages. cpu by default.
-TORCH_NIGHTLY_URL="https://download.pytorch.org/whl/nightly/cpu"
-
-# If cuda is available, use the pytorch nightly with cuda for faster execution on cuda GPUs.
-test -f /usr/bin/nvidia-smi && TORCH_NIGHTLY_URL="https://download.pytorch.org/whl/nightly/cu121"
+# If cuda is available, based on presence of nvidia-smi, install the pytorch nightly
+# with cuda for faster execution on cuda GPUs.
+if [[ -x "$(command -v nvidia-smi)" ]]; 
+then
+  TORCH_NIGHTLY_URL="https://download.pytorch.org/whl/nightly/cu121"
+else
+  TORCH_NIGHTLY_URL="https://download.pytorch.org/whl/nightly/cpu"
+fi
 
 # pip packages needed by exir.
 REQUIREMENTS_TO_INSTALL=(