sleepy/Gemma4Pad
1
0
Fork
You've already forked Gemma4Pad
 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

[app] Full Gemma4Pad iPadOS implementation #4

Merged
sleepy merged 2 commits from feature/1-full-implementation into main 2026-05-22 12:59:12 +02:00
Conversation 0 Commits 2 Files changed 20 +2007
sleepy commented 2026-05-22 12:45:56 +02:00
Owner
Copy link

Complete iPadOS app for on-device Gemma 4 inference.

What

  • SwiftUI app with NavigationSplitView (sidebar + detail)
  • MLX Swift inference via gemma-4-swift-mlx (Gemma4Pipeline)
  • Text, image (PhotosPicker), and audio (AVAudioEngine) input
  • Multi-session conversation management with JSON persistence
  • Model selection: Gemma 4 E2B/E4B with Q4/Q8 quantization
  • HuggingFace model download via Gemma4Pipeline
  • Web search integration via configurable SearXNG endpoint
  • Editable system prompt with search tool schema toggle
  • Light/dark mode
  • XcodeGen project.yml + setup script

Architecture

  • Framework: MLX Swift (Apple's preferred LLM framework per WWDC 2025)
  • Inference: gemma-4-swift-mlx library (text + vision + audio)
  • Models: Pre-quantized MLX models from mlx-community on HuggingFace
  • Storage: Conversations in JSON files, attachments in Documents dir

Files (20 changed, ~1988 lines)

Closes

  • #1 (project scaffold)
  • #2 (services layer)
  • #3 (UI layer)
Complete iPadOS app for on-device Gemma 4 inference. ## What - SwiftUI app with NavigationSplitView (sidebar + detail) - MLX Swift inference via gemma-4-swift-mlx (Gemma4Pipeline) - Text, image (PhotosPicker), and audio (AVAudioEngine) input - Multi-session conversation management with JSON persistence - Model selection: Gemma 4 E2B/E4B with Q4/Q8 quantization - HuggingFace model download via Gemma4Pipeline - Web search integration via configurable SearXNG endpoint - Editable system prompt with search tool schema toggle - Light/dark mode - XcodeGen project.yml + setup script ## Architecture - **Framework**: MLX Swift (Apple's preferred LLM framework per WWDC 2025) - **Inference**: gemma-4-swift-mlx library (text + vision + audio) - **Models**: Pre-quantized MLX models from mlx-community on HuggingFace - **Storage**: Conversations in JSON files, attachments in Documents dir ## Files (20 changed, ~1988 lines) ## Closes - #1 (project scaffold) - #2 (services layer) - #3 (UI layer)
- SwiftUI app with NavigationSplitView (sidebar + detail)
- MLX Swift inference via gemma-4-swift-mlx (Gemma4Pipeline)
- Text, image (PhotosPicker), and audio (AVAudioEngine) input
- Multi-session conversation management with JSON persistence
- Model selection: Gemma 4 E2B/E4B with Q4/Q8 quantization
- HuggingFace model download via Gemma4Pipeline
- Web search integration via configurable SearXNG endpoint
- Editable system prompt with search tool schema toggle
- Light/dark mode support
- XcodeGen project.yml + setup script

Closes #1, #2, #3
- Fix prompt double-formatting: use pipeline multi-turn API properly
- Add NSMicrophoneUsageDescription to Info.plist
- Fix Swift 6 @Sendable violation in progress callback
- Remove dead code from ModelManager, use cache directory only
- Async image loading in MessageBubble (no sync I/O on main thread)
- Single source of truth for recording state (AudioRecorder)
- Throw on unknown model instead of silent fallback
- Remove print() in production SessionManager
- Strip HTML tags in WebSearchService content fetch
sleepy merged commit 6698b5a35e into main 2026-05-22 12:59:12 +02:00
Sign in to join this conversation.
Reviewers
No reviewers
Labels
Clear labels
No items
No labels
Milestone
Clear milestone
No items
No milestone
Projects
Clear projects
No items
No project
Assignees
Clear assignees
No assignees
1 participant
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set.

Reference
sleepy/Gemma4Pad!4
No description provided.