ARPL makes llama.cpp adapt to ARM ISA and core topology at runtime
OpeningTough145 · reddit · 2026-08-04
ARPL is a public project that configures llama.cpp based on the actual ARM chip it is running on, instead of using one generic setup for every device.
What it detects at runtime
- Available ISA extensions such as SDOT, I8MM, and SME2.
- CPU core clustering / topology.
- It then adjusts thread counts and context parameters automatically.
What the repo includes
- An Android reference app in Kotlin/Compose with a JNI bridge to llama.cpp.
- HWCAP-based runtime ISA detection.
- Topology-aware thread count recommendations.
- Context patches for flash attention and KV-cache quantization.
Status
- Built and tested on a Samsung S25 Ultra.
- Heterogeneous CPU/GPU/NPU partitioning is not included yet.
- The release is noncommercial under PolyForm Noncommercial.
Related event: ARPL Enables Runtime Adaptation for llama.cpp on ARM Devices(2 posts)→
More from Embodied
- NVIDIA’s 14B DreamZero model passes 30 simulated trials but fails can-to-mug tasks — wandb · 2026-08-04
- Concept mockup imagines a self-driving minivan with fold-out beds — LukeW · 2026-08-04
- Multimodal spatial intelligence and physical AI pitched as the next frontier — richdotca · 2026-08-04
- Pure VLAs may not need long-horizon planning if VLMs can cover it — m_wulfmeier · 2026-08-04
- Agility Robotics’ early Digit research robots helped seed China’s humanoid boom — chris_j_paxton · 2026-08-04
- Super Mario 64 demo on Vision Pro turns the screen into a portal — pvncher · 2026-08-04