Q&A: How to Submit Large AI/ML AppImages to AppImageHub?

Summary

I'm trying to submit TalkType (an AI-powered speech recognition application) to AppImageHub, but the CI test times out after 10 minutes due to the AppImage size (887MB). The AppImage requires PyTorch for local AI inference, which accounts for most of the size.

Details

Project: TalkType - AI-powered speech-to-text dictation for Linux Wayland
AppImage Size: 887MB (optimized as much as possible)
PR: AppImage/appimage.github.io#3547
GitHub Release: https://github.com/ronb1964/TalkType/releases/tag/v0.3.7

What the AppImage includes:

• Python 3.11 runtime
• PyTorch with CUDA support (for GPU acceleration)
• faster-whisper AI models
• GTK3 for system tray
• All dependencies fully bundled

Size breakdown:

• PyTorch CUDA library: 861MB (libtorch_cuda.so)
• PyTorch CPU library: 422MB (libtorch_cpu.so)
• These are core PyTorch components required for AI inference

Testing status:

✅ AppImage works perfectly on multiple distributions (Fedora, Ubuntu-based)
✅ Desktop file validates
✅ AppStream metadata included
✅ All GTK dependencies bundled
✅ Thoroughly tested in both CPU and GPU modes

The problem:

The CI workflow has a hardcoded 10-minute timeout (line 20 in .github/workflows/test.yml):

timeout-minutes: 10

The download + extraction + testing of an 887MB AppImage exceeds this limit.

Questions

1. Is there a recommended approach for large AI/ML applications on AppImageHub?

   • Should we host the AppImage somewhere with faster download speeds?
   • Can the timeout be extended for certain categories of applications?

2. Are there other AI/ML applications on AppImageHub?

   • I found Upscayl (AI image upscaler) in the database, but couldn't determine its size or how it passed CI.

3. Is it possible to skip the CI test for verified applications?

   • The AppImage has been thoroughly tested and works correctly.
   • Would manual review by a maintainer be an alternative?

Why this matters

As AI/ML applications become more common on Linux, AppImageHub will likely see more submissions that bundle ML frameworks like PyTorch, TensorFlow, or ONNX Runtime. These frameworks are inherently large but necessary for local, privacy-focused AI applications.

Additional context

The AppImage is already available on GitHub releases and working perfectly for users. AppImageHub inclusion would increase discoverability and provide users with a trusted source for Linux applications.

Any guidance would be greatly appreciated!

Relevant links:

• PR: Add TalkType - AI-powered speech-to-text for Linux Wayland appimage.github.io#3547
• Project: https://github.com/ronb1964/TalkType
• Release: https://github.com/ronb1964/TalkType/releases/tag/v0.3.7