model: /tmp/e001-june/qwen25_monolithic_full_lut4.mlmodelc  function: infer
total ops: 6411
  none   4242  (66.2%)
  ANE   2090  (32.6%)
  CPU     79  (1.2%)

=== assigned-compute ops only (ANE+CPU+GPU = 2169) ===
  ANE  2090  (96.4% of compute)
  CPU    79  (3.6% of compute)
  GPU     0  (0.0% of compute)

=== top op types in 'none' ===
  const 4028
  ios18.constexpr_lut_to_dense 213
  identity 1

=== top op types in 'ANE' ===
  ios18.mul 338
  ios18.concat 225
  ios18.slice_by_index 225
  ios18.conv 212
  ios18.reshape 170
  ios18.transpose 170
  ios18.add 141
  ios18.expand_dims 113

=== top op types in 'CPU' ===
  ios18.cast 19
  ios18.reduce_argmax 16
  ios18.gather_along_axis 16
  ios18.transpose 15
  ios18.squeeze 4
  ios18.add 2
  ios18.concat 2
  select 2
KV-machinery placement  (model: /tmp/e001-june/qwen25_monolithic_full_lut4.mlmodelc  fn:infer)
  slice_update ANE=56
  read_state ANE=57
  write_state ANE=56
  slice_by_index ANE=225
  gather ANE=2  CPU=1
  gather_along_axis CPU=16