Hi, I’ve tested the model with depthwise conv on CPU using autotvm, such error appeared:
Tuning…
Traceback (most recent call last):
File “tune_relay_x86_dwconv.py”, line 190, in
tune_and_evaluate(tuning_option)
File “tune_relay_x86_dwconv.py”, line 164, in tune_and_evaluate
tune_kernels(tasks, **tuning_opt)
File “tune_relay_x86_dwconv.py”, line 126, in tune_kernels
target=target, template_key=‘direct’)
File “/root/tvm/python/tvm/autotvm/task/task.py”, line 191, in create
sch, _ = func(*args)
File “/root/tvm/topi/python/topi/x86/depthwise_conv2d.py”, line 217, in _topi_nn_depthwise_conv2d_NCHWc
data_layout, out_layout, dtype)
File “/root/tvm/topi/python/topi/x86/depthwise_conv2d.py”, line 78, in _depthwise_conv2d_NCHWc_cpu
= get_const_tuple(kernel.shape)
ValueError: not enough values to unpack (expected 6, got 4)
input_shape = (1, 8, 7, 7)
data = mx.symbol.Variable(name="data")
sym = mx.sym.Convolution(data=data, num_filter=8, kernel=(3,3), num_group=8, stride=(1,1), pad=(1,1), name='conv1')
net, params = relay.frontend.from_mxnet(sym, shape={'data': input_shape}, dtype=dtype)TVM version: 0.6.dev