Issue with Tuning Kernels for ResNet-18 on x86 CPU Using TVM

Troubleshooting
#1

I’ve been following the tutorial here: https://web.tvm.apache.org/docs/how_to/tune_with_autotvm/tune_relay_x86.html to auto-tune a convolutional network for x86 CPUs, but I ran into an issue during the graph-level tuning phase.

The error I’m getting is: Config for target=llvm -keys=cpu, workload=('dense_nopack.x86', ('TENSOR', (1, 512), 'float32'), ('TENSOR', (1000, 512), 'float32'), None, 'float32') is missing in ApplyGraphBest context. A fallback configuration is used, which may bring great performance regression.

This happens when evaluating the tuned model at the graph level, resulting in only a slight improvement over the kernel-level tuning. The fallback configuration seems to be causing performance regression.

Here’s what I’ve tried so far:

  1. Verified the log_file and graph_opt_sch_file paths.
  2. Checked that all workloads were extracted and included during tuning.
  3. Adjusted the number and repeat parameters in autotvm.LocalRunner.

Despite these steps, the issue persists. Has anyone encountered this problem? How can I ensure the correct configurations are applied to avoid fallback scenarios during graph-level tuning?

Any advice would be greatly appreciated.