Update install docs and add troubleshooting page (#21210)

docs/genai/howto/build-model.md

@@ -4,7 +4,7 @@ description: How to build models with ONNX Runtime generate() API
 has_children: false
 parent: How to
 grand_parent: Generate API (Preview)
-nav_order: 2
+nav_order: 3
 ---
 
 # Generate models using Model Builder

docs/genai/howto/install.md

@@ -13,28 +13,58 @@ nav_order: 1
 * TOC placeholder
 {:toc}
 
+## Pre-requisites
+
+### CUDA
+
+If you are installing the CUDA variant of onnxruntime-genai, the CUDA toolkit must be installed.
+
+The CUDA toolkit can be downloaded from the [CUDA Toolkit Archive](https://developer.nvidia.com/cuda-toolkit-archive).
+
+Ensure that the `CUDA_PATH` environment variable is set to the location of your CUDA installation.
+
 ## Python packages
 
+Note: only one of these packages should be installed in your application.
+
+### CPU
+
 ```bash
 pip install numpy
 pip install onnxruntime-genai --pre
 ```
+
+### DirectML
+
 Append `-directml` for the library that is optimized for DirectML on Windows
 
 ```bash
 pip install numpy
 pip install onnxruntime-genai-directml --pre
 ```
 
+### CUDA
+
 Append `-cuda` for the library that is optimized for CUDA environments
 
+#### CUDA 11
+
 ```bash
 pip install numpy
 pip install onnxruntime-genai-cuda --pre --index-url=https://aiinfra.pkgs.visualstudio.com/PublicPackages/_packaging/onnxruntime-genai/pypi/simple/
 ```
 
+#### CUDA 12
+
+```bash
+pip install numpy
+pip install onnxruntime-genai-cuda --pre --index-url=https://aiinfra.pkgs.visualstudio.com/PublicPackages/_packaging/onnxruntime-cuda-12/pypi/simple/
+```
+
 ## Nuget packages
 
+Note: only one of these packages should be installed in your application.
+
 ```bash
 dotnet add package Microsoft.ML.OnnxRuntimeGenAI --prerelease
 ```

docs/genai/howto/troubleshoot.md

@@ -0,0 +1,34 @@
+---
+title: Troubleshoot
+description: How to troubleshoot common problems
+has_children: false
+parent: How to
+grand_parent: Generate API (Preview)
+nav_order: 4
+---
+
+# Troubleshoot issues with ONNX Runtime generate() API
+{: .no_toc }
+
+* TOC placeholder
+{:toc}
+
+## Installation issues
+
+### Windows Conda import error
+
+```
+ImportError: DLL load failed while importing onnxruntime_genai: A dynamic link library (DLL) initialization routine failed.
+```
+
+If you see this issue in a Conda environment on Windows, you need to upgrade the `C++ runtime for Visual Studio`. In the conda environment, run the following command:
+
+```bash
+conda install conda-forge::vs2015_runtime
+```
+
+The onnxruntime-genai Python package should run without error after this extra step.
+
+### Windows CUDA import error
+
+After CUDA toolkit installation completed on windows, ensure that the `CUDA_PATH` system environment variable has been set to the path where the toolkit was installed. This variable will be used when importing the onnxruntime_genai python module on Windows. Unset or incorrectly set `CUDA_PATH` variable may lead to a `DLL load failed while importing onnxruntime_genai`.

docs/genai/tutorials/phi3-v.md

@@ -62,17 +62,28 @@ Support for Windows machines with GPUs other than NVIDIA is coming soon!
    ```
    This command downloads the model into a folder called `cuda-int4-rtn-block-32`.
 
-2. Install the generate() API
+2. Setup your CUDA environment
 
-   ```
+   Install the [CUDA toolkit](https://developer.nvidia.com/cuda-toolkit-archive).
+
+   Ensure that the `CUDA_PATH` environment variable is set to the location of your CUDA installation.
+
+
+3. Install the generate() API
+
+   * CUDA 11
+
+   ```bash
    pip install numpy
    pip install --pre onnxruntime-genai-cuda --index-url=https://aiinfra.pkgs.visualstudio.com/PublicPackages/_packaging/onnxruntime-genai/pypi/simple/
    ```
 
-3. Setup your CUDA environment
-
-   Please follow the steps [here](../howto/setup-cuda-env.md) to setup the CUDA environment.
+   * CUDA 12
 
+   ```bash
+   pip install numpy
+   pip install onnxruntime-genai-cuda --pre --index-url=https://aiinfra.pkgs.visualstudio.com/PublicPackages/_packaging/onnxruntime-cuda-12/pypi/simple/
+   ```
 
 4. Run the model