ARM dotprod instructions unsupported in old system

[a65tesla]

I have a Samsung S9+ Snapdragon smartphone. When I try to launch any model, it loads into RAM, but when the response is generated, the application crashes. I tried different models, including Gemma 4.

Gemma 4 works great in the Layla AI app, but I don't like Layla's interface, so I decided to switch to this app, but unfortunately it doesn't work for me. What could be the problem?

Here are my phone specs
Processor: Snap 845
RAM 6GB
ROM 128GB UFS 2.1

I also tried it on Mi8 Pro, but it had the same crashes.

I use the model gemma-4-E2B-it-IQ4_NL.gguf

[a65tesla]

The Gemma 4 model itself is a new turboQuant model, And it can work on devices with 8GB of RAM and even on 6GB of RAM it can work

[Spike8086]

Found the root cause. The current build uses ARM dotprod instructions which require Snapdragon 855 (2019) or newer. Your Snapdragon 845 doesn't support these instructions — the model loads into RAM fine, but crashes when inference starts.
I would fix it in future release(not be too soon,it's a work need a lots of efforts, maybe need 1-2 months, i would release other new version during this process, not means it get solved, but once it get solved,i would add a new comment in here), thanks for your understandings and considerations.If you wanna use CloverPal immeadiately, v1.0.0 still fit for your devices(but not support gemma4/qwen3.5)

[a65tesla]

Thanks for the reply! I'll wait for the update! Good luck with your work!