Feature Request: ARMv7 / Termux Support on Mobile Devices

[BrianxBorne]

Prerequisites

• I am running the latest code. Mention the version if possible as well.
• I carefully followed the README.md.
• I searched using keywords relevant to my issue to make sure that I am creating a new issue that is not already open (or closed).
• I reviewed the Discussions, and have a new and useful enhancement to share.

Feature Description

Hi and thank you for the incredible work on llama.cpp.

I’ve been trying to build and run the project on an ARMv7 Android device via Termux. While compilation begins successfully with LLAMA_PORTABLE=1 , it ultimately fails due to architecture-specific issues like NEON intrinsic redefinitions and other compatibility errors. Even when the build partially succeeds, the resulting binaries don’t run reliably.

It seems that full support for ARMv7 and Termux environments may not be currently in scope. Just wanted to raise this in case broader compatibility with mobile devices is considered in the future. Thanks again for the amazing project

Motivation

I believe adding support—or at least reducing the barrier—for Termux/ARMv7 devices could unlock a significant set of use cases, especially in regions where mobile devices are the primary computing platform. These devices are often affordable, widely available, and increasingly capable of basic on-device inference for small models like TinyLLaMA.

Having a portable, lightweight LLM that runs entirely offline on such devices would have huge implications for accessibility, privacy-preserving AI, education, and localized chatbots in low-resource settings. The ability to run even a tiny GGUF model locally on mobile without internet could be transformative.

I understand this may be outside current priorities, but I wanted to share the value this feature could bring and express appreciation for the project. Thank you again for the amazing tool you've created and continue to improve.

Possible Implementation

No response

[techindev]

So where are you going, do you want to create a server type? , oh something like that?

[baonudesifeizhai]

So where are you going, do you want to create a server type? , oh something like that?

maybe package to .so

[BrianxBorne]

So where are you going, do you want to create a server type? , oh something like that?

I'm trying to get llama.cpp running entirely offline in Termux on an ARMv7 Android device no server, just a self-contained CLI chatbot using something like TinyLLaMA.

Here's a screenshot of part of the CMake build output on my device.

As you can see, the CPU is detected as armv7l, and many of the machine-specific tests fail — especially ones involving NEON, i8mm, and SVE instructions, which ARMv7 doesn't support. Even pthread detection is a bit flaky.

I’ve been trying to work around this by forcing LLAMA_PORTABLE=1 and disabling certain flags, but some things still cause runtime issues.

If anyone has suggestions or alternative approaches to improve compatibility on minimal hardware like this, it would be a huge help.