inceptionv3 for VKNN — fp16, full-GPU
Inception v3 is a multi-branch classifier built on factorized convolutions (~27M parameters). 299x299 RGB in, 1000 ImageNet class scores out.
Inception v3 image classifier (torchvision export), compiled for the VKNN Vulkan inference engine. Every op runs on the GPU (fp16 weights + fp16 compute, fp32 host boundary); the CPU backend serves as the bit-accuracy oracle in VKNN's device gates, and this model's GPU outputs are gated against an fp32 onnxruntime golden (cosine > 0.999, argmax agreement) on-device.
| File | inceptionv3_fp16.vxm (46M) |
| Input | input [1x3x299x299] (NCHW fp32) |
| Container | VXM3 (single graph, weights embedded) |
Run
# any VKNN example binary, e.g. the IO runner:
./vknn_run_io inceptionv3_fp16.vxm out_dir input.bin
Recompile from your own ONNX with vknn_compile model.onnx out.vxm --fp16.