Assistance Needed: FramePack Installation Issues on Windows 11 with RTX 5070Ti (PyTorch Dependency Conflicts)

Nobody has claimed this yet.

Dear FramePack Author,

I hope this message finds you well. I am reaching out for assistance with installing and running FramePack on Windows 11 using an RTX 5070Ti GPU. I have encountered several dependency-related issues that are preventing FramePack from utilizing GPU acceleration, causing the "VAE encoding" stage to run very slowly on the CPU. I have detailed the problems, steps I’ve tried, and my questions below, hoping you can provide some guidance.

---

### Background
I am working in the project directory `C:\Users\JXin\Documents\framepack_cu126_torch26` and have set up a virtual environment `framepack_env` (Python 3.10). My goal is to get FramePack running with GPU acceleration on my RTX 5070Ti. However, I’ve faced multiple issues with PyTorch dependencies and other requirements.

---

### Issues Encountered

#### 1. PyTorch 2.7.0+cu121 Wheel Not Found
- An engineer suggested using PyTorch 2.7.0+cu121 for Windows (CUDA 12.1), as CUDA 12.8 (`cu128`) is only supported on Linux. I tried installing it with:
pip install torch==2.7.0+cu121 torchvision==0.17.0+cu121 -f https://download.pytorch.org/whl/cu121
- Error received:
ERROR: Could not find a version that satisfies the requirement torch==2.7.0+cu121 (from versions: 1.11.0, ..., 2.6.0)
ERROR: No matching distribution found for torch==2.7.0+cu121
- I couldn’t find `torch-2.7.0+cu121-cp310-cp310-win_amd64.whl` in the specified wheel directory.

#### 2. Downgraded to PyTorch 2.6.0+cu126, but torchvision-0.16.0+cu126 Not Found
- Since 2.7.0+cu121 was unavailable, I downgraded to PyTorch 2.6.0+cu126 (suggested by my project directory name `framepack_cu126_torch26` and known to be stable). I believe `torch-2.6.0+cu126-cp310-cp310-win_amd64.whl` exists, but I couldn’t find `torchvision-0.16.0+cu126-cp310-cp310-win_amd64.whl` at `https://download.pytorch.org/whl/cu126`.
- Newer versions like `torchvision-0.21.0+cu126` are available, but they don’t match PyTorch 2.6.0.
- I am hesitant to install `torchvision==0.16.0` directly from PyPI (without CUDA specification), as I’m unsure if this will prevent FramePack from using the GPU.

#### 3. SageAttention Detection Issue
- A user mentioned that FramePack might require SageAttention, but the provided wheel (`sageattention-2.1.1+cu128torch2.7.0-cp310-cp310-win_amd64.whl`) is for PyTorch 2.7.0+cu128, which isn’t compatible with Windows.
- They noted that even after installing SageAttention, FramePack fails to detect it (showing "not installed"). I have skipped this for now, but I’m unsure if SageAttention is a required dependency for FramePack.

#### 4. NumPy Version Conflict
- Different sources suggest different NumPy versions for FramePack:
- An engineer recommended NumPy >= 2.2.0 (I installed 2.2.3).
- A successful case used `numpy==1.26.2` with PyTorch 2.6.0.
- I’m concerned that the NumPy version might cause compatibility issues with FramePack or other dependencies like SageAttention.

#### 5. Current Status
- I have not yet successfully installed all dependencies, so FramePack is still running on the CPU, making the "VAE encoding" stage very slow.
- A successful case reported that FramePack worked on an RTX 5070Ti (Windows 11) after installing PyTorch, torchvision, torchaudio, `numpy==1.26.2`, and `flatbuffers`, then running `run.bat` or `update.bat

Start by reviewing the Windows setup path through run.bat and update.bat, then reproduce the reported PyTorch, torchvision, NumPy, and SageAttention installation combinations. Check whether the resulting environment reaches GPU-backed VAE encoding; done means FramePack installs successfully and uses the RTX 5070Ti rather than CPU execution.

Written by the indexing model from the issue text.