[Bug]: recent models that crash the app

[Disonantemus]

Describe the bug
I found some recent Models that crash the app:

• falcon3-7b-instruct (tested: iq4_xs)
• gemma3-4b-it (tested: q2_k)
• ibm-granite_granite-3.2-8b-instruct (tested: iq4_xs, q4_0)
• phi-4-mini-instruct (tested: iq2_m)

They are incompatible?
Maybe because they are tools section in Ollama? (Structured output, function calling)

To Reproduce

Steps to reproduce the behavior:

1. Download any quantization from the model
2. Select model
3. App crash and close

Expected behavior

To not crash and start chat.

Screenshots

Impossible to do.

Environment (please complete the following information):

• Device: Samsung SM-G780G
• OS: Android 13
• App Version: 1.8.5 (Last)

Additional context

All recommended models works, and everything else that I've tried from Huggingface, I've not tried with vision or audio models.

[TFWol]

On iOS, the same models don't crash the app, but they will fail to load.

[Disonantemus]

Describe the bug
I found some recent Models that crash the app:

• falcon3-7b-instruct (tested: iq4_xs)
• gemma3-4b-it (tested: q2_k)
• ibm-granite_granite-3.2-8b-instruct (tested: iq4_xs, q4_0)
• phi-4-mini-instruct (tested: iq2_m)

All this models working now with last version: Version 1.8.8 (58) [Android].

This bug seems related to llama-cpp not been in sync with latest.

[a-ghorbani]

see: #238