[Bug] v0.0.94 very slow

[DragonIRQ]

[Off Grid v0.0.94] This update is terrible and very slow; the previous update was much better.

[dishit-wednesday]

hey @DragonIRQ ,would like to understand it better ,
what are u referring to here , is it ui rendering ,navigation screens or in terms of
model streaming responses .

[DragonIRQ]

The program is generally slow, especially when importing or running a model, or it crashes. Previously, in the earlier version, it didn't suffer from these issues.

[dishit-wednesday]

Hey ,from the last github release,not much has changed for android it was mostly an ios focused update .
so not sure why this experience on your side .It’s running fine on my test devices currently.

Could you share which device you’re using and, if possible, a bit more detail on where you notice the slowdown or crashes
(model loading, importing, generation, UI navigation, etc.)?
Will look into it .

[098orin]

I’m experiencing the same issue. Aside from screen scrolling, almost every important part of the app becomes slow — button clicks, model responses, and basically everything else I can think of.

[dragonballirq]

Exactly the old version is much better, I don't know why. ..I'm currently using version 89, which is much better than version 94.

[dishit-wednesday]

understood thanks for reporting this everyone ,as of now I am unable to point to a specific thing since there have been no changes for the android side in the last 2 releases ,but I will investigate this more and get it fixed by the next release .
look into ui optimization .
just curious everyone have you guys installed the apk from github or someone has tried the playstore release as well ??

[my-garbage]

Yeah it got slower. Its really half of the tokens now. I thought I just would have a bad Dax but if others experience that too it seems to he an issue.

[dishit-wednesday]

Hey guys, I'm trying to look into this issue and gather a bit more information.

Since a few of you have mentioned that both the UI and model responses seem slower, I'd like to check a couple of things:

• Which backend is currently selected in Model Settings?
• Had any of you previously switched to the HTP/NPU backend back in v0.0.88?

In Advanced settings, does the text below the backend buttons say "Offloading layers to Hexagon NPU" even though nothing looks selected?

Could you also try the following?

1. Go to Model Settings and switch the backend to CPU.
2. If your context length is set very high, try reducing it as well.
3. After making those changes, go back to the chat screen and try sending a message.
4. A reload banner should appear there since the model settings changed. Please reload the session when prompted.
5. Let me know if that helps or improves performance.

• Does this happen with all models, or mainly specific model families such as Qwen?

Thanks for helping investigate this.

[my-garbage]

Yeah same settings here.

[my-garbage]

As it was said there was nothing changed.

Image generation also doesn't work anymore for me.

My device Tensor G5 might not support the model please select a CPU model.
I used that model chillout_mix before.

BTW ... These permanent error windows could be copyable.

[dishit-wednesday]

Thanks for checking and sharing the details.

It's a bit difficult to pinpoint exactly what's going wrong because OffGrid intentionally doesn't collect usage analytics or logs in order to preserve user privacy.

To help investigate, I've created a special APK that captures application lifecycle and diagnostic logs (startup, model loading, backend selection, crashes, etc.).

It does not collect chats, prompts, model data, personal information, or other app content. The logs stay stored locally on your device and are never shared unless you explicitly choose to share them. Nothing is sent to me automatically, and you'll be able to review the logs yourself before sharing anything.

If any of you are willing to try it, here's the APK:

https://drive.google.com/file/d/1ELIDIkZteWvr_-hkjCPHm_2WUYUYkt0V/view?usp=sharing
You don't need to do anything extensive. Just use the app normally for a minute or two:

• Navigate through a few screens
• Eject/load a model
• Send a simple prompt
• Reproduce any slowdown, lag, or crash if possible

After that, use the Share button, upload the log file to Google Drive (or any file-sharing service), and send me the file/link here.

@my-garbage if you're open to it, could you also try reproducing the image generation/loading error you mentioned? The logs should help me understand what's failing there as well.

No pressure if you'd rather not. I'll continue investigating on my end too, but logs from an affected device would make it much easier to identify the root cause.

Thanks a lot for helping track this down.

[dishit-wednesday]

[my-garbage]

The Error of picture generation is gone with the update to the special version

[dishit-wednesday]

is it ? thats strange but great .
if u were able to use it a bit could u share me the logs ,would like to see whats causing the ui issue ,if u are comfortable sharing after auditing the logs