Bonsai-4B โ 1-bit for the Qualcomm Hexagon NPU (HNPU)
A QHexRT-runnable bundle of prism-ml/Bonsai-4B at true 1-bit (Q1_0, ~1.125 bit/weight) for the Hexagon NPU. Runs the full model on-device via a custom HVX op-package.
Supported arches: v75 (SM8650), v81 (SM8850). Artifacts are arch-pinned; this repo contains v75/ and v81/ directories.