Issues Installing PyTorch 2.7.0 and torchvision 0.17.0 with CUDA 12.9 on Windows 11 - Missing Wheel Files and CUDA Not Enabled

[kaworukevin]

Hello everyone,

I'm trying to install PyTorch 2.7.0 and torchvision 0.17.0 with CUDA support on Windows 11, but I'm facing several issues. I need to align with the versions used by my team’s engineers and ultimately run a project called FramePack. I’d really appreciate your help!

My Environment

Operating System: Windows 11 Pro x64

Python Version: 3.10.6

GPU: NVIDIA GeForce RTX 5070 Ti

NVIDIA Driver Version: 576.02

CUDA Version: 12.9 (as shown by nvidia-smi)

Virtual Environment: C:\Users\JXin\Documents\framepack_cu126_torch26\venv

My Goal

Install PyTorch 2.7.0 and torchvision 0.17.0 with CUDA 12.8 support (compatible with CUDA 12.9).

Ensure PyTorch enables CUDA (currently getting "Torch not compiled with CUDA enabled").

Install Triton 3.3.0 and SageAttention 2.1.1 (need Windows-compatible wheel files).

Resolve the NumPy initialization warning: UserWarning: Failed to initialize NumPy: _ARRAY_API not found.

What I've Done

Environment Cleanup:

Removed all WSL2 and Ubuntu residuals (used wsl --unregister and deleted ext4.vhdx).

Cleaned the system-level Python environment (C:\Users\JXin\AppData\Local\Programs\Python\Python310), removing all torch and nvidia-* related packages.

Created a virtual environment: python -m venv venv.

Attempted PyTorch Installation:

Used the command:

pip install torch==2.7.0 torchaudio==2.7.0 --index-url https://download.pytorch.org/whl/cu128 --no-cache-dir
pip install torchvision==0.17.0 --no-cache-dir

Result: Installed CPU versions (2.7.0+cpu, 0.17.0+cpu), not the expected 2.7.0+cu128.

Manual Wheel File Download:

Downloaded:

torch-2.7.0+cu128-cp310-cp310-win_amd64.whl

torchaudio-2.7.0+cu128-cp310-cp310-win_amd64.whl

Installed:

pip install C:\Users\JXin\Downloads\torch-2.7.0+cu128-cp310-cp310-win_amd64.whl
pip install C:\Users\JXin\Downloads\torchaudio-2.7.0+cu128-cp310-cp310-win_amd64.whl

Verified:

python -c "import torch; print(torch.version); print(torch.cuda.is_available())"

Output:

2.7.0+cu128
True

Confirmed PyTorch now supports CUDA.

Issue: Could Not Find torchvision Wheel File:

Could not find torchvision-0.17.0+cu128-cp310-cp310-win_amd64.whl at https://download.pytorch.org/whl/cu128.

Tried using pip to download automatically:

pip install torchvision==0.17.0 --index-url https://download.pytorch.org/whl/cu128 --no-cache-dir

Result: Still installed CPU version (0.17.0+cpu).

NumPy Warning:

During verification, I get:

UserWarning: Failed to initialize NumPy: _ARRAY_API not found

Current NumPy version might not be fully compatible with PyTorch 2.7.0.

My Questions

Cannot Find torchvision-0.17.0+cu128-cp310-cp310-win_amd64.whl:

Does anyone know where I can download this wheel file, or if there’s an alternative version (e.g., 0.18.0 or 0.16.0) that’s compatible with PyTorch 2.7.0?

If using pip to download automatically, do I need to adjust the --index-url?

NumPy Warning:

How can I resolve UserWarning: Failed to initialize NumPy: _ARRAY_API not found? Should I upgrade NumPy to a specific version (e.g., 2.0.0)?

SageAttention Wheel File:

I need sageattention-2.1.1+cu128torch2.7.0+cp310+cp310-win_amd64.whl (Windows version).

I only found sageattention-2.1.1+cu126torch2.6.0-cp310-cp310-win_amd64.whl (for PyTorch 2.6.0 and CUDA 12.6). Is this compatible? If not, where can I find the correct version?

What I've Tried

Verified CUDA environment:

nvidia-smi shows CUDA 12.9, driver is working.

Environment variable CUDA_PATH is set to C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.9.

Tried alternative torchvision versions (e.g., 0.18.0 and 0.16.0), but I'm concerned about compatibility with PyTorch 2.7.0.

Upgraded NumPy:

pip install --force-reinstall numpy==2.0.0

But the warning persists.

Request for Help

How can I resolve the issue of not finding torchvision-0.17.0+cu128-cp310-cp310-win_amd64.whl?

Is there another way to install PyTorch 2.7.0 and torchvision 0.17.0 with CUDA 12.8/12.9 on Windows 11?

How can I fix the NumPy warning?

Where can I find the correct SageAttention 2.1.1 wheel file?

You can also reach me directly at kaworukevin@gmail.com if you have any suggestions or files to share.

Thanks for your help!

[TheRealElroso]

Why the torchvision-0.17.0+cu128-cp310-win_amd64.whl Doesn’t Exist

Official wheels for PyTorch 2.7/CUDA 12.8 (“+cu128”) were released only for Linux (x86_64 & arm64). The Windows build pipeline is still on CUDA 12.1 for stable releases. Hence:

torch-2.7.0+cu128-cp310-win_amd64.whl = not published

torchvision-0.17.0+cu128-cp310-win_amd64.whl = not published

What you found (torch-2.7.0+cpu / torchvision-0.17.0+cpu) is the fallback CPU wheel.

Nightly Windows wheels with CUDA 12.8 are likewise not yet on the public index (only Linux). See the PyTorch 2.7 release post and “Previous Versions” matrix.

---

Practical Ways Forward on Windows 11

---

Fixing the NumPy “_ARRAY_API not found” Warning

PyTorch 2.7 was compiled against NumPy 2.2 API.
Install any NumPy ≥ 2.2 (2.3.0 is the current stable):

pip install --upgrade "numpy>=2.2,<3.0"

That suppresses the _ARRAY_API warning.

---

TL;DR for Kevin

• There are no official CUDA 12.8/+cu128 wheels for Windows yet.
• Use torch/torchvision/torchaudio 2.7.0 +cu121 on Windows, or run the Linux wheels under WSL2/cloud.
• NumPy warning → upgrade NumPy ≥ 2.2.
• SageAttention: either stick with Torch 2.6 + cu126 wheel or compile from source.

Hope that clears it up!

[kaworukevin]

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

Dear [Forum/Support Team],

I am reaching out for assistance with setting up FramePack on Windows 11 using an RTX 5070Ti GPU (CUDA-compatible). Currently, the "VAE encoding" stage of FramePack runs very slowly because it is using the CPU, and I want to enable GPU acceleration. I have created a virtual environment framepack_env (Python 3.10) in my project directory C:\Users\JXin\Documents\framepack_cu126_torch26, but I am facing several issues while installing PyTorch and its dependencies. Below is a summary of the problems I’ve encountered:

---

Issues Encountered

1. Cannot Find PyTorch 2.7.0+cu121 Wheel

• I tried installing PyTorch 2.7.0+cu121 as recommended by an engineer (since it is supposed to be supported on Windows). Command used:
  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
• An engineer mentioned that PyTorch 2.7.0 with CUDA 12.8 (cu128) is only supported on Linux, not Windows, and I should use CUDA 12.1 (cu121) for Windows. However, I still cannot find torch-2.7.0+cu121-cp310-cp310-win_amd64.whl.

2. Cannot Find torchvision-0.16.0+cu126 Wheel

• Since 2.7.0+cu121 was not found, I downgraded to PyTorch 2.6.0+cu126 (as my project directory name framepack_cu126_torch26 suggests this version, and 2.6.0 is a known working version). I believe torch-2.6.0+cu126-cp310-cp310-win_amd64.whl exists, but I cannot find torchvision-0.16.0+cu126-cp310-cp310-win_amd64.whl.
• I checked https://download.pytorch.org/whl/cu126 and found newer versions like torchvision-0.21.0+cu126, but 0.16.0+cu126 seems to be missing.
• I am considering installing torchvision==0.16.0 directly from PyPI (without CUDA specification), but I am worried this might prevent FramePack from using the GPU.

3. CUDA 12.8 Not Supported on Windows

• I downloaded torch-2.7.0+cu128-cp310-cp310-win_amd64.whl, but the engineer confirmed that PyTorch 2.7.0 with CUDA 12.8 is only for Linux, not Windows. This explains why I cannot use cu128 wheels on Windows.

4. SageAttention Issue (Possibly Required by FramePack)

• Another response mentioned that FramePack might require SageAttention. They provided a wheel (sageattention-2.1.1+cu128torch2.7.0-cp310-cp310-win_amd64.whl), but it is for PyTorch 2.7.0+cu128, which is not compatible with Windows.
• Even if SageAttention is installed, FramePack might not detect it (showing "not installed"). I have skipped this for now, but I am concerned it might be necessary for FramePack to work properly.

5. NumPy Version Conflict

• FramePack requires NumPy, but different responses suggest different versions:
  • An engineer recommended NumPy >= 2.2.0 (I installed 2.2.3).
  • A successful case (using PyTorch 2.6.0) used numpy==1.26.2.
• I am worried that the NumPy version might cause compatibility issues with FramePack or other dependencies like SageAttention.

6. Current Status of FramePack

• I have not yet successfully installed all dependencies, so FramePack is still running on the CPU (making "VAE encoding" very slow).
• A successful case suggested running run.bat or update.bat in the framepack_cu126_torch26 directory and ensuring flatbuffers is installed, but I haven’t reached this step yet.

---

Steps I Have Tried

1. Created a virtual environment framepack_env (Python 3.10).
2. Attempted to install PyTorch 2.7.0+cu121, but the wheel was not found.
3. Downgraded to PyTorch 2.6.0+cu126, but could not find torchvision-0.16.0+cu126.
4. Confirmed that CUDA 12.8 is not supported on Windows; I should use CUDA 12.1 or 12.6.
5. Installed numpy==1.26.2 and flatbuffers (as per the successful case).

---

My Questions

1. How can I find or replace torchvision-0.16.0+cu126-cp310-cp310-win_amd64.whl to work with PyTorch 2.6.0+cu126?
2. If torchvision-0.16.0+cu126 is unavailable, can I use pip install torchvision==0.16.0 directly? Will this affect FramePack’s GPU acceleration?
3. Is SageAttention a required dependency for FramePack? If so, how can I find a version compatible with PyTorch 2.6.0+cu126?
4. Will the NumPy version (1.26.2 vs. 2.2.3) cause conflicts? Which version should I use?
5. Are there alternative methods to get FramePack running with GPU acceleration on Windows 11 using an RTX 5070Ti?

---

Environment Details

• Operating System: Windows 11
• GPU: NVIDIA RTX 5070Ti
• Python Version: 3.10
• Project Directory: C:\Users\JXin\Documents\framepack_cu126_torch26
• Virtual Environment: framepack_env

---

I would greatly appreciate any help or suggestions to resolve these issues. Please feel free to reach out to me at kaworukevin@gmail.com if you need more details.

Thank you for your time and assistance!

Best regards,
Kevin

[woct0rdho]

@TheRealElroso No, torch 2.7.0+cu128 wheels are released for Windows.

@kaworukevin You can download torchvision from https://download.pytorch.org/whl/cu128/torchvision-0.22.0%2Bcu128-cp310-cp310-win_amd64.whl

Or you can just run

pip install -U torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu128

where -U means upgrade and pip will automatically find the latest versions. Without -U, pip may not upgrade already installed versions.

If you see any error, please show how you run this command and all output results.

[chirusandeep]

Hi @kaworukevin

Env used: conda

Run for Env:

conda create --name framepack python=3.12
conda activate framepack

Python: 3.12.x

python --version

Run:

pip install --pre torch torchvision torchaudio --index-url https://download.pytorch.org/whl/nightly/cu128

If encountered issues run pip uninstall torch and run again torch install cmd.

Execute:

pip install -r requirements.txt
pip install sageattention
python demo_gradio.py

My specs :

i7-12Gen, RTX 4070 Laptop GPU 8gb VRAM and 64GB RAM, M.2 NVME 1TB

Default settings image to video generation:
Per second video == 2~2.5 Minutes Generation TIme