Run yolov5m6.onnx meets error

1

Hello,

I run yolov5m6.onnx got from export yolov5 to onnx, meet CUDA_ERROR_LAUNCH_OUT_OF_RESOURCES, when running:

m = graph_executor.GraphModule(lib["default"](dev))

for item in inputs_format:
    m.set_input('images', tvm.nd.array(input_data.astype(numpy.float32)))

m.run()

after:

with tvm.transform.PassContext(opt_level=2):
    lib = relay.build(mod, target=target, params=params)

But strange, yolov5l6.onnx model can run. Can anyone explain this?

I run on 16GB nvidia T4 GPU. Without search & tunning.