[SOLVED] [AutoTVM] Tuning log


#1


[Task 30/31] Current/Best: 2.45/ 7.44 GFLOPS | Progress: (536/1000) | 760.70 s Done.
[Task 31/31] Current/Best: 2.02/ 2.25 GFLOPS | Progress: (480/1000) | 645.60 s Done.
Compile…
WARNING:autotvm:Cannot find config for target=llvm -device=arm_cpu -target=arm-linux-gnueabihf -mattr=+neon, workload=(‘conv2d_transpose_nchw’, (1, 64, 3, 3, ‘float32’), (64, 64, 2, 2, ‘float32’), (2, 2), (0, 0), ‘float32’). A fallback configuration is used, which may bring great performance regression.
WARNING:autotvm:Cannot find config for target=llvm -device=arm_cpu -target=arm-linux-gnueabihf -mattr=+neon, workload=(‘conv2d_transpose_nchw’, (1, 64, 6, 6, ‘float32’), (64, 64, 2, 2, ‘float32’), (2, 2), (0, 0), ‘float32’). A fallback configuration is used, which may bring great performance regression.
Upload…
Evaluate inference time cost…
Mean inference time (std dev): 34.75 ms (0.25 ms)

Is there anyone can help me? Thanks in advance for any answer.@merrymercy


#2

If you are using the out-of-the-box NNVM autotvm tuning scripts, you might need to add the conv2d transpose operator to the operator types to tune. Try adding nnvm.sym.conv2d_transpose to this argument.


#3

It works, thanks!

# /tvm/tutorials/autotvm/tune_nnvm_arm.py
tasks = autotvm.task.extract_from_graph(net, target=target,
                                     shape={'data': input_shape}, dtype=dtype,
                                     symbols=(nnvm.sym.conv2d,nnvm.sym.conv2d_transpose,))