Installation of Python and Dependencies.md

Here is a comprehensive, step-by-step installation guide created by synthesizing the official Patreon post (`Installation.md`) and the detailed transcript (`Installation Transcript using Old Versions.md`).

This guide provides the _what_ from the new post and the _how-and-why_ from the old transcript, giving you a complete and robust process.

---

## AI/ML Environment Setup Guide for Windows

This guide details the proper installation and configuration of all necessary components for a stable AI development environment on Windows, based on the latest recommended versions.

**Installation Order:** The recommended installation order is:

1. Visual Studio Build Tools
2. Visual Studio Community Edition
3. CUDA
4. cuDNN
5. TensorRT
6. Python
7. FFmpeg, Git, and other tools
8. System Configuration


---

### 1. Visual Studio Build Tools

This component is required for compiling Python libraries (wheels) on your local machine.

1. **Download:** Use the official "v14 latest" link for version 17.14.13: [vs_BuildTools.exe Download](https://download.visualstudio.microsoft.com/download/pr/89ee1303-1ba2-4f66-b07c-5099983fd1e4/a5fcbc55bd633d6c79c9319317a8cb77c70227ce87b70948c27649f7384c2909/vs_BuildTools.exe).

2. **Install:** Run the installer.

3. **Select Workload:** In the installer, check the box for **"Desktop development with C++"**.

4. **Components:** On the right-hand panel, ensure all components are selected, including the Windows SDKs.

5. Complete the installation.


### 2. Visual Studio Community Edition

This provides the MSVC compiler tools, which are also required by some libraries.

1. **Download:** Get the "Community" edition from the official page: [Visual Studio Downloads](https://visualstudio.microsoft.com/downloads/).
2. **Install:** Run the installer and select the same **"Desktop development with C++"** workload.
3. **Components:** Select all components _except_ "Incredibuild - Build Acceleration".
4. **Add to PATH (Critical):** After installation, you must manually add the compiler to your system's Environment Variables.

    - Find the cl.exe path. It will be similar to:
        C:\Program Files\Microsoft Visual Studio\2022\Community\VC\Tools\MSVC\14.44.35207\bin\Hostx64\x64
    - Open "Edit the system environment variables".
    - Click "Environment Variables...".
    - Under "System variables," find and "Edit..." the **`Path`** variable.
    - Click "New" and paste the full path you found above.
    - Click OK on all windows.

1. **Verification:** Open a **new** PowerShell (not CMD) window and type `cl`. You should see the "Microsoft (R) C/C++ Optimizing Compiler" message, not an error.

### 3. CUDA Installation

This is the core NVIDIA toolkit for GPU operations. We will install version 12.9.1.

1. **Download:** [CUDA 12.9.1 Download (Windows)](https://developer.download.nvidia.com/compute/cuda/12.9.1/local_installers/cuda_12.9.1_576.57_windows.exe).
2. **Install:** Run the executable.
3. **Select Options:**
    - Agree and Continue.
    - Choose **Custom (Advanced)** installation.
    - **Uncheck** components you do not need, such as "GeForce Experience". Ensure "CUDA" and "Driver components" are selected.
4. **Paths:** Let it install to the default `C:\Program Files\...` locations.
5. **Restart:** Restart your PC after installation.
6. **Verification:** Open a **new** CMD window and type `nvcc --version`. You must see the 12.9.1 version listed. If you see an error, it means the `PATH` variables were not set correctly.


### 4. cuDNN Installation

This library accelerates deep neural network operations and is critical for performance.

1. **Download:** Get the cuDNN 9.12.0 for CUDA 12.9 zip file.
    - Hugging Face (Backup): [cudnn_9.12_cuda_12.9.zip](https://huggingface.co/MonsterMMORPG/Wan_GGUF/resolve/main/cudnn_9.12_cuda_12.9.zip)
    - Official Archive (Requires login): [NVIDIA cuDNN Archive](https://developer.nvidia.com/cudnn-archive)
2. **Install:** This is a **manual** process.
    - The `Installation.md` document implies an installer that places files in `C:\Program Files\NVIDIA\CUDNN`. If you download the zip, you must extract it.
    - Extract the contents of the zip file.
    - Copy the extracted `v9.12` folder into `C:\Program Files\NVIDIA\CUDNN\` (you may need to create the `CUDNN` folder). Your final structure should look like `C:\Program Files\NVIDIA\CUDNN\v9.12`.
3. **Add to PATH:** You must add two paths to your System `Path` variable:
    - `C:\Program Files\NVIDIA\CUDNN\v9.12\bin\12.9`
    - `C:\Program Files\NVIDIA\CUDNN\v9.12`
4. **Add CUDNN_PATH:** You must also create a _new_ System Variable.
    - **Variable name:** `CUDNN_PATH`
    - **Variable value:** `C:\Program Files\NVIDIA\CUDNN\v9.12\bin\12.9`

### 5. TensorRT Installation

This library optimizes models for inference speed.

1. **Download:** Get the TensorRT 10.13.2 zip file for CUDA 12.9.
    - Hugging Face (Backup): [TensorRT-10.13.2.6...zip](https://huggingface.co/MonsterMMORPG/Wan_GGUF/resolve/main/TensorRT-10.13.2.6.Windows.win10.cuda-12.9.zip?download=true)
2. **Install:** This is also a manual extraction.
    - Extract the zip file directly to your C drive, resulting in a path like `C:\TensorRT_10_13_2_6`.
3. **Add to PATH:** Add the following two paths to your System `Path` variable:
    - `C:\TensorRT_10_13_2_6\lib`
    - `C:\TensorRT_10_13_2_6`

### 6. Python Installation

We will install Python 3.10.11 as the default, as it is the most widely supported version.

1. **Download:** [Python 3.10.11 64-bit Installer](https://www.python.org/ftp/python/3.10.11/python-3.10.11-amd64.exe).
2. **Install:** Run the installer.
3. **Critical Setup Steps:**
    - Check the box **"Add Python.exe to PATH"**.
    - Select **"Customize installation"**.
    - On the next screen, check all "Optional Features" boxes.
    - On the "Advanced Options" screen, check all boxes (especially "py launcher").
    - **Change the install location** to a simple path, like `C:\Python310`.
4. **Final Step:** After installation, a "Setup was successful" screen will appear. Click the button that says **"Disable path length limit"**. This is very important.
5. **Verification:** Open a **new** CMD window and type `python --version`. You should see "Python 3.10.11".

> **Note on Other Versions:** Some applications may require other Python versions like 3.11.9 or 3.12.10. The provided 1-click installers are designed to automatically create virtual environments (venvs) with the correct Python version if they detect it's installed on your system. For now, only 3.10.11 is required.

### 7. Other Essential Tools

#### FFmpeg

Used for video processing in many AI applications.

1. **Download:** Use the specific linked version: [ffmpeg...win64-gpl.zip](https://huggingface.co/MonsterMMORPG/Wan_GGUF/resolve/main/ffmpeg-N-121105-ga0936b9769-win64-gpl.zip).
2. **Install:**
    - Create a new folder: `C:\ffmpeg_latest`.
    - Extract the zip file.
    - Go into the extracted folder, find the `bin` directory.
    - Copy the contents of the `bin` directory (e.g., `ffmpeg.exe`, `ffprobe.exe`) and paste them directly into your `C:\ffmpeg_latest` folder.
3. **Add to PATH:** Add `C:\ffmpeg_latest` to your System `Path` variable.
4. **Verification:** Open a **new** CMD window and type `ffmpeg`. You should see the FFmpeg version information.

#### Git

Used for cloning (downloading) code from repositories.

1. **Download:** [Git for Windows](https://git-scm.com/downloads).
2. **Install:** Run the installer. Click "Next" on all prompts, accepting all default settings.
3. **Add to PATH (Manual):** The installer should do this, but it often fails. Manually add these two paths to your System `Path` variable to prevent errors:
    - `C:\Program Files\Git\cmd`
    - `C:\Program Files\Git\bin`
4. **Verification:** Open a **new** CMD window and type `git`. You should see the Git help message.

#### NodeJS

Required by some applications.

1. **Download:** [NodeJS 22.20 (LTS)](https://nodejs.org/dist/v22.20.0/node-v22.20.0-x64.msi).
2. **Install:** Run the installer and accept all defaults.

---

### 8. System Configuration

#### Enable Windows Long Paths

Prevents errors with deep folder structures in AI projects.

1. **Download:** [Enable_Windows_Long_Paths.reg](https://www.patreon.com/file?h=111553210&m=550698799).
2. **Run:** Double-click the downloaded `.reg` file and accept the prompt to modify your registry.

#### Virtual RAM (Page File)

Crucial for loading large models, which can use system RAM (and virtual RAM) as overflow.

1. Set up a **minimum of 100 GB** (102400 MB) of virtual RAM (page file) on your fastest drive (e.g., your C: drive if it's an NVMe SSD).
2. **How-to:** Search for "View advanced system settings" > Performance "Settings..." > "Advanced" tab > Virtual memory "Change..." > Uncheck "Automatically manage..." > Select C: drive > "Custom size" > set Initial and Maximum to `102400`.

---

### 9. Environment Variables Final Check

After all this, your System `Path` variable should contain (at minimum) these entries. The order matters, with Python, cuDNN, and your main CUDA version near the top.

- `C:\Python310\Scripts\`
- `C:\Python310\`
- `C:\Program Files\NVIDIA\CUDNN\v9.12\bin\12.9`
- `C:\Program Files\NVIDIA\CUDNN\v9.12`
- `C:\Program Files\NVIDIA\CUDNN\v9.12\bin\12.9` (This is the `CUDNN_PATH` value, also good to have in `Path`)
- `C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.9\bin`
- `C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.9\libnvvp`
- `C:\TensorRT_10_13_2_6\lib`
- `C:\TensorRT_10_13_2_6`
- `C:\ffmpeg_latest`
- `C:\Program Files\Git\cmd`
- `C:\Program Files\Git\bin`
- `C:\Program Files\Microsoft Visual Studio\2022\Community\VC\Tools\MSVC\14.44.35207\bin\Hostx64\x64`
- ... (other Windows system paths)


---

### 10. Common Troubleshooting Fixes

This is where the old transcript shines. If an application fails, it's almost always a `PATH` or DLL issue.

#### **Problem:** `git` is not recognized...

You see an error `git is not recognized as an internal or external command`.

- **Fix:** The installer failed to add Git to your `PATH`. Follow step 7 ("Git") above to **manually add** `C:\Program Files\Git\cmd` and `C:\Program Files\Git\bin` to your System `Path` variable. Close and re-open CMD to check.


#### **Problem:** `CUDA_PATH is set but CUDA wasn't able to be loaded` (cuDNN Mismatch)

An application (like `LivePortrait` in the transcript) fails to load, complaining about CUDA or cuDNN. This means the application was built for a _different_ cuDNN version (e.g., 8.9) than your system default (9.12).

- **Fix (The "Override" Method):**

    1. Go to the [NVIDIA cuDNN Archive](https://developer.nvidia.com/cudnn-archive) and download the **zip file** for the specific version the app needs (e.g., `cuDNN 8.9.7`).
    2. Create a new, simple folder (e.g., `C:\cudnn8.9`).
    3. Extract the zip's contents (the `bin`, `include`, `lib` folders) into `C:\cudnn8.9`.
    4. Go to your System `Path` variables.
    5. Click "New" and add the path to the `bin` folder of this new version (e.g., `C:\cudnn8.9\bin`).
    6. **Crucially:** Use the "Move Up" button to move this new entry to the **very top** of the `Path` list.
    7. This makes it the _new default_ for your system, overriding the 9.12 version. Run your app. Remember to move it back down when you're done.

#### **Problem:** `ONNX Runtime` Fails (DLL Hell)

An application (like `FaceFusion` in the transcript) fails, and the console shows errors related to `onnxruntime` and missing DLLs. This means the app's isolated environment can't find the necessary cuDNN files.

- **Fix (The "Manual Copy" Method):**

    1. Go to your `cuDNN` installation folder (the one the app _needs_, e.g., `C:\Program Files\NVIDIA\CUDNN\v9.12`).
    2. Open the `bin\12.9` folder and copy all the `.dll` files.
    3. Navigate to the app's Python virtual environment (venv). The path will be something like:
        C:\FaceFusion_v3\venv\Lib\site-packages\onnxruntime\
    4. Paste all the `.dll` files you copied directly into this `onnxruntime` folder. This manually provides the app with the libraries it was missing.

#### **Problem:** `Triton` Errors

You get errors related to `triton` during model loading or training.

- **Fix:** This is often a caching issue.

    1. Go to your user folder: `C:\Users\[YourUsername]\`
    2. Delete the entire `.triton` folder.
    3. A `bat` file to do this automatically is also provided here: [Clear_Triton_Cache.zip](https://www.patreon.com/file?h=111553210&m=526715040).