Roadmap for SmolChat

[shubham0204]

Features To Be Implemented

• Long-press chat message to reveal actions (copy, share, edit)
  • Blocker: onLongClick does not register on ChatMessageText, neither via Modifier nor TextView.setOnLongClickListener
• Button to copy code snippets
  • Blocker: noties-markdown does not support this natively. We might have to tweak the Markdown parser or implement our code.
• Add top_k and top_p sampling parameters
  • Blocker: Leads to unknown native crashes during conversations, needs debugging
• Fix state management in ChatActivity to preserve screen state after configuration changes (like screen rotation)
• Support vision models
• Add TTS support for messages
• Support attaching documents to the chat (RAG)
• Show model downloads within the app's UI (and not the system's DownloadManager)
• Add 'reset to default settings' button in the settings screen
• Add Jinja template parsing
• 'Stop' button in chat should not clear the last incomplete message
• For context length usage, show another label besides RAM usage, instead of a separate dialog
• Add speech-to-text capabilities for entering queries

[shubham0204]

Long-press chat message to reveal actions (copy, share, edit)
Blocker: onLongClick does not register on ChatMessageText, neither via Modifier nor TextView.setOnLongClickListener

Implemented in commit 27a0ba1

[Namitjain07]

i think u should also include:

1. better resource manager(user flexibility)
2. a much modern UI implementation

[anna-hope]

Do you have any plans/desire to allow uploading images, so we could use computer vision models for things like OCR? Or would that fall under

Support attaching documents to the chat (RAG)

[shubham0204]

@anna-hope Supporting vision models in SmolChat is in the plan, but that would require making changes in the C++ wrapper class that the app uses on top of llama.cpp. This effort will take some time, hence we have no upper bound on the time this feature might take to complete.