LFM2.5-24B-A2B MoE on Qualcomm IQ9075 NPU via KernelX
We ran Liquid AI’s LFM2.5-24B-A2B MoE fully on the Qualcomm IQ-9075 NPUs. KernelX-tuned kernels deliver up to 1.50× faster prefill and 1.40× faster decode than upstream llama.cpp.
We ran Liquid AI’s LFM2.5-24B-A2B MoE fully on the Qualcomm IQ-9075 NPUs. KernelX-tuned kernels deliver up to 1.50× faster prefill and 1.40× faster decode than upstream llama.cpp.