Transposing via --channel_format and --input_shapes fails

[masroof-maindak]

Hello.

My original ONNX model was NCHW.

My goal is to get rid of client side image-transposing by making the transpose operation a part of the bmodel itself.

We have a yolov8m based model, and the following snippet has been modified from the official sophon-demo yolov8_plus_det app. It works perfectly, with no issues.

model_transform.py \
        --model_name $model_name \
        --model_def $onnx_path \
        --input_shapes [[$1,3,352,640]] \
        --mean 0.0,0.0,0.0 \
        --scale 0.0039216,0.0039216,0.0039216 \
        --keep_aspect_ratio \
        --pixel_format rgb \
        --mlir ${model_name}_$1b.mlir

The problem arises when I try to transpose (notice --input_shapes and --channel_format:

    model_transform.py \
        --model_name $model_name \
        --model_def $onnx_path \
        --input_shapes [[$1,352,640,3]] \
        --channel_format nhwc \
        --mean 0.0,0.0,0.0 \
        --scale 0.0039216,0.0039216,0.0039216 \
        --keep_aspect_ratio \
        --pixel_format rgb \
        --mlir ${model_name}_$1b.mlir

In this case, the MLIR output gets generated just fine, but the model_deploy script fails.

model_deploy.py \
        --mlir ${model_name}_$1b.mlir \
        --quantize F32 \
        --chip $target \
        --model ${model_name}_fp32_$1b.bmodel

[Running]: tpuc-opt yolov8m_bm1684x_f32_final.mlir --codegen="model_file=yolov8m_fp32_1b.bmodel embed_debug_info=False model_version=latest bmodel_only=False gdma_check=True" -o /dev/null
The dir path of compiler_profile is "./"
bmcpu init: skip cpu_user_defined
Cannot open libusercpu.so, disable user cpu layer.
size does not match for tensor model.0.conv.weight
readTensor failed
UNREACHABLE executed at /__w/tpu-mlir/tpu-mlir/lib/Support/TensorFile.cpp:293!
PLEASE submit a bug report to https://github.com/llvm/llvm-project/issues/ and include the crash backtrace.
Stack dump:
0.      Program arguments: tpuc-opt yolov8m_bm1684x_f32_final.mlir --init "--codegen=model_file=yolov8m_fp32_1b.bmodel embed_debug_info=False model_version=latest bmodel_only=False gdma_check=True" --deinit --mlir-print-debuginfo -o /dev/null
 #0 0x00005651f4152ff7 (/usr/local/lib/python3.10/dist-packages/tpu_mlir/bin/tpuc-opt+0x91cff7)
 #1 0x00005651f4150d1e (/usr/local/lib/python3.10/dist-packages/tpu_mlir/bin/tpuc-opt+0x91ad1e)
 #2 0x00005651f415397a (/usr/local/lib/python3.10/dist-packages/tpu_mlir/bin/tpuc-opt+0x91d97a)
 #3 0x00007fd3ceb85520 (/usr/local/lib/python3.10/dist-packages/tpu_mlir/lib/third_party/libc.so.6+0x42520)
 #4 0x00007fd3cebd99fc pthread_kill (/usr/local/lib/python3.10/dist-packages/tpu_mlir/lib/third_party/libc.so.6+0x969fc)
 #5 0x00007fd3ceb85476 gsignal (/usr/local/lib/python3.10/dist-packages/tpu_mlir/lib/third_party/libc.so.6+0x42476)
 #6 0x00007fd3ceb6b7f3 abort (/usr/local/lib/python3.10/dist-packages/tpu_mlir/lib/third_party/libc.so.6+0x287f3)
 #7 0x00005651f4150b41 (/usr/local/lib/python3.10/dist-packages/tpu_mlir/bin/tpuc-opt+0x91ab41)
 #8 0x00005651f5b8a2c2 (/usr/local/lib/python3.10/dist-packages/tpu_mlir/bin/tpuc-opt+0x23542c2)
 #9 0x00005651f5b89dfa (/usr/local/lib/python3.10/dist-packages/tpu_mlir/bin/tpuc-opt+0x2353dfa)
#10 0x00005651f5908c9d (/usr/local/lib/python3.10/dist-packages/tpu_mlir/bin/tpuc-opt+0x20d2c9d)
#11 0x00005651f59021f7 (/usr/local/lib/python3.10/dist-packages/tpu_mlir/bin/tpuc-opt+0x20cc1f7)
#12 0x00005651f4b95a4e (/usr/local/lib/python3.10/dist-packages/tpu_mlir/bin/tpuc-opt+0x135fa4e)
#13 0x00005651f42a534e (/usr/local/lib/python3.10/dist-packages/tpu_mlir/bin/tpuc-opt+0xa6f34e)
#14 0x00005651f4b8fb08 (/usr/local/lib/python3.10/dist-packages/tpu_mlir/bin/tpuc-opt+0x1359b08)
#15 0x00005651f4b8dc5b (/usr/local/lib/python3.10/dist-packages/tpu_mlir/bin/tpuc-opt+0x1357c5b)
#16 0x00005651f4b8a93e (/usr/local/lib/python3.10/dist-packages/tpu_mlir/bin/tpuc-opt+0x135493e)
#17 0x00005651f5a5db04 (/usr/local/lib/python3.10/dist-packages/tpu_mlir/bin/tpuc-opt+0x2227b04)
#18 0x00005651f5a5e131 (/usr/local/lib/python3.10/dist-packages/tpu_mlir/bin/tpuc-opt+0x2228131)
#19 0x00005651f5a605d8 (/usr/local/lib/python3.10/dist-packages/tpu_mlir/bin/tpuc-opt+0x222a5d8)
#20 0x00005651f414467b (/usr/local/lib/python3.10/dist-packages/tpu_mlir/bin/tpuc-opt+0x90e67b)
#21 0x00005651f4143a44 (/usr/local/lib/python3.10/dist-packages/tpu_mlir/bin/tpuc-opt+0x90da44)
#22 0x00005651f5c7b058 (/usr/local/lib/python3.10/dist-packages/tpu_mlir/bin/tpuc-opt+0x2445058)
#23 0x00005651f413dd4a (/usr/local/lib/python3.10/dist-packages/tpu_mlir/bin/tpuc-opt+0x907d4a)
#24 0x00005651f413e214 (/usr/local/lib/python3.10/dist-packages/tpu_mlir/bin/tpuc-opt+0x908214)
#25 0x00005651f413cc5a (/usr/local/lib/python3.10/dist-packages/tpu_mlir/bin/tpuc-opt+0x906c5a)
#26 0x00007fd3ceb6cd90 (/usr/local/lib/python3.10/dist-packages/tpu_mlir/lib/third_party/libc.so.6+0x29d90)
#27 0x00007fd3ceb6ce40 __libc_start_main (/usr/local/lib/python3.10/dist-packages/tpu_mlir/lib/third_party/libc.so.6+0x29e40)
#28 0x00005651f413c065 (/usr/local/lib/python3.10/dist-packages/tpu_mlir/bin/tpuc-opt+0x906065)
Aborted (core dumped)
Traceback (most recent call last):
  File "/usr/local/lib/python3.10/dist-packages/tpu_mlir/python/tools/model_deploy.py", line 678, in <module>
    tpu_patterns = tool.build_model()
  File "/usr/local/lib/python3.10/dist-packages/tpu_mlir/python/tools/model_deploy.py", line 402, in build_model
    patterns = mlir_to_model(
  File "/usr/local/lib/python3.10/dist-packages/tpu_mlir/python/utils/mlir_shell.py", line 1066, in mlir_to_model
    _os_system(cmd, log_level=log_level)
  File "/usr/local/lib/python3.10/dist-packages/tpu_mlir/python/utils/mlir_shell.py", line 495, in _os_system
    raise RuntimeError("[!Error]: {}".format(cmd_str))
RuntimeError: [!Error]: tpuc-opt yolov8m_bm1684x_f32_final.mlir --codegen="model_file=yolov8m_fp32_1b.bmodel embed_debug_info=False model_version=latest bmodel_only=False gdma_check=True" -o /dev/null

What am I doing wrong, or is there any other way to transpose an image from within the model, without resorting to having the client transpose the image before sending?

[masroof-maindak]

As per the new 'Bug Report' issue instructions in 572c8e6:

My chip is a BM1684x, and here is the Dockerfile I am using to spin-up a tpu-mlir container:

FROM sophgo/tpuc_dev:latest

RUN wget https://github.com/sophgo/tpu-mlir/releases/download/v1.27/tpu_mlir-1.27-py3-none-any.whl && \
    pip install tpu_mlir-1.27-py3-none-any.whl[onnx]

Source model: ONNX Yolov8m

Tpu-mlir version: Summary: Machine learning compiler based on MLIR for TPU v1.27-g6979325-20260206

[masroof-maindak]

Additionally, I should mention that I have also tried using the --fuse_preprocess and --customization_format arguments, but to no avail, and get the same error as before:

#!/bin/bash
# Note that these models will accept uint8 NHWC images as input
model_dir=$(dirname $(readlink -f "$0"))
target=bm1684x
target_dir=BM1684X

outdir=../models/$target_dir
function gen_mlir() {
    onnx_path=../yolov8m/best.onnx

    model_transform.py \
        --model_name $model_name \
        --model_def $onnx_path \
        --input_shapes [[$1,352,640,3]] \
        --channel_format nhwc \
        --input_types uint8 \
        --mean 0.0,0.0,0.0 \
        --scale 0.0039216,0.0039216,0.0039216 \
        --keep_aspect_ratio \
        --pixel_format rgb \
        --mlir ${model_name}_$1b.mlir
}

function gen_fp32bmodel() {
    model_deploy.py \
        --mlir ${model_name}_$1b.mlir \
        --quantize F32 \
        --chip $target \
        --fuse_preprocess \
        --customization_format RGB_PACKED \
        --model ${model_name}_fp32_$1b.bmodel

    mv ${model_name}_fp32_$1b.bmodel $outdir/
}

# function gen_fp16bmodel() {
#     model_deploy.py \
#         --mlir ${model_name}_$1b.mlir \
#         --quantize F16 \
#         --chip $target \
#         --model ${model_name}_fp16_$1b.bmodel

#     mv ${model_name}_fp16_$1b.bmodel $outdir/
# }

mkdir -p $outdir

# batch_size=1
model_name=yolov8m
gen_mlir 1
gen_fp32bmodel 1
# gen_fp16bmodel 1

---

Here is the full log log for the above script:

root@3a2fc527d8ca:/workspace/workdir# ./gen_fp32bmodel_mlir.sh
2026/04/28 15:51:21 - INFO : TPU-MLIR v1.27-20260206
2026/04/28 15:51:21 - INFO : 
         _____________________________________________________ 
        | preprocess:                                           |
        |   (x - mean) * scale                                  |
        '-------------------------------------------------------'
  config Preprocess args : 
        resize_dims           : same to net input dims
        keep_aspect_ratio     : True
        keep_ratio_mode       : letterbox
        pad_value             : 0
        pad_type              : center
        --------------------------
        mean                  : [0.0, 0.0, 0.0]
        scale                 : [0.0039216, 0.0039216, 0.0039216]
        --------------------------
        pixel_format          : rgb
        channel_format        : nhwc
        yuv_type              : 

2026/04/28 15:51:24 - INFO : Input_shape assigned
2026/04/28 15:51:24 - WARNING : ConstantFolding failed.
2026/04/28 15:51:24 - INFO : ConstantFolding finished
2026/04/28 15:51:24 - INFO : skip_fuse_bn:False
2026/04/28 15:51:25 - INFO : Onnxsim opt finished
2026/04/28 15:51:25 - WARNING : ConstantFolding failed.
2026/04/28 15:51:25 - INFO : ConstantFolding finished
2026/04/28 15:51:25 - INFO : Save mlir file: yolov8m_1b_origin.mlir
[Running]: tpuc-opt yolov8m_1b_origin.mlir --struct-optimize --shape-infer --canonicalize --extra-optimize -o yolov8m_1b.mlir
[Success]: tpuc-opt yolov8m_1b_origin.mlir --struct-optimize --shape-infer --canonicalize --extra-optimize -o yolov8m_1b.mlir
2026/04/28 15:51:26 - INFO : Mlir file generated:yolov8m_1b.mlir
2026/04/28 15:51:27 - INFO : TPU-MLIR v1.27-20260206
[Success] Create config file 'yolov8m_bm1684x_f32.layer_group_config.json'.
Content:
{
  "shape_secs_search_strategy": 0,
  "structure_detect_opt": true,
  "sc_method_configs": [
    {
      "sc_method": "sc_method_quick_search",
      "MAX_TRY_NUM": 20
    },
    {
      "sc_method": "sc_method_search_better_v1",
      "NSECS_SEARCH_RECORD_THRESHOLD": 3,
      "CSECS_SEARCH_RECORD_THRESHOLD": 3,
      "DSECS_SEARCH_RECORD_THRESHOLD": 3,
      "HSECS_SEARCH_RECORD_THRESHOLD": 3,
      "WSECS_SEARCH_RECORD_THRESHOLD": 3
    },
    {
      "sc_method": "sc_method_search_better_v2",
      "MAX_NSECS": 32,
      "MAX_CSECS": 32,
      "MAX_DSECS": 32,
      "MAX_HSECS": 32,
      "MAX_WSECS": 32,
      "NSECS_SEARCH_RECORD_THRESHOLD": 2,
      "CSECS_SEARCH_RECORD_THRESHOLD": 2,
      "DSECS_SEARCH_RECORD_THRESHOLD": 2,
      "HSECS_SEARCH_RECORD_THRESHOLD": 2,
      "WSECS_SEARCH_RECORD_THRESHOLD": 2
    }
  ]
}
[Running]: tpuc-opt yolov8m_1b.mlir --processor-assign="chip=bm1684x mode=F32 num_device=1 num_core=1 addr_mode=auto high_precision=False" --processor-top-optimize --fuse-preprocess="mode=F32 customization_format=RGB_PACKED align=False" --convert-top-to-tpu="weightFileName=yolov8m_bm1684x_f32_tpu_weights.npz asymmetric=False doWinograd=False q_group_size=0 q_symmetric=False matmul_perchannel=False gelu_mode=normal" --canonicalize --weight-fold -o yolov8m_bm1684x_f32_tpu.mlir
The dir path of compiler_profile is "./"
bmcpu init: skip cpu_user_defined
Cannot open libusercpu.so, disable user cpu layer.
Entering FusePreprocessPass.
Inserting PermuteOp.
Inserting SliceOp.
Inserting DWConvOp.
Inserting SwapChannelOp.
[Success]: tpuc-opt yolov8m_1b.mlir --processor-assign="chip=bm1684x mode=F32 num_device=1 num_core=1 addr_mode=auto high_precision=False" --processor-top-optimize --fuse-preprocess="mode=F32 customization_format=RGB_PACKED align=False" --convert-top-to-tpu="weightFileName=yolov8m_bm1684x_f32_tpu_weights.npz asymmetric=False doWinograd=False q_group_size=0 q_symmetric=False matmul_perchannel=False gelu_mode=normal" --canonicalize --weight-fold -o yolov8m_bm1684x_f32_tpu.mlir
[Running]: tpuc-opt yolov8m_bm1684x_f32_tpu.mlir --mlir-disable-threading --strip-io-quant="quant_input=False quant_output=False quant_input_list= quant_output_list= quant_output_bf16=False quant_input_int8=False quant_output_int8=False " --processor-tpu-optimize --dev-parallel --weight-reorder  --subnet-divide="dynamic=False" --op-reorder --topo-sort --layer-group="opt=2 group_by_cores=auto compress_mode=none debugger=0 disable_group_overlap=false lgcache=true config_filename= enable_lghash=False lghash_dir="  --core-parallel  --after-layergroup-weight-reorder --address-assign -o yolov8m_bm1684x_f32_final.mlir --debug_cmd=
The dir path of compiler_profile is "./"
bmcpu init: skip cpu_user_defined
Cannot open libusercpu.so, disable user cpu layer.
Dumping LgConfig!
Global Configs:
  shape_secs_search_strategy = 0
  structure_detect_opt = 1
Search Method Config: sc_method_quick_search
  CSECS_SEARCH_RECORD_THRESHOLD = -1
  DSECS_SEARCH_RECORD_THRESHOLD = -1
  HSECS_SEARCH_RECORD_THRESHOLD = -1
  MAX_CSECS = 32
  MAX_DSECS = 32
  MAX_HSECS = 32
  MAX_NSECS = 32
  MAX_TRY_NUM = 20
  MAX_WSECS = 32
  NSECS_SEARCH_RECORD_THRESHOLD = -1
  WSECS_SEARCH_RECORD_THRESHOLD = -1
Search Method Config: sc_method_search_better_v1
  CSECS_SEARCH_RECORD_THRESHOLD = 3
  DSECS_SEARCH_RECORD_THRESHOLD = 3
  HSECS_SEARCH_RECORD_THRESHOLD = 3
  MAX_CSECS = 32
  MAX_DSECS = 32
  MAX_HSECS = 32
  MAX_NSECS = 32
  MAX_TRY_NUM = 20
  MAX_WSECS = 32
  NSECS_SEARCH_RECORD_THRESHOLD = 3
  WSECS_SEARCH_RECORD_THRESHOLD = 3
Search Method Config: sc_method_search_better_v2
  CSECS_SEARCH_RECORD_THRESHOLD = 2
  DSECS_SEARCH_RECORD_THRESHOLD = 2
  HSECS_SEARCH_RECORD_THRESHOLD = 2
  MAX_CSECS = 32
  MAX_DSECS = 32
  MAX_HSECS = 32
  MAX_NSECS = 32
  MAX_TRY_NUM = 20
  MAX_WSECS = 32
  NSECS_SEARCH_RECORD_THRESHOLD = 2
  WSECS_SEARCH_RECORD_THRESHOLD = 2
Load debugger file "yolov8m_bm1684x_f32.layer_group_config.json" success!
output_bytes: 112560, before group, NetStatisticPass total_bytes: 128967576
==---------------------------==
Run LayerGroupSearchPass : 
    Searching the optimal layer groups
==---------------------------==
Failed to open hash file from all available paths:
  - current directory: 15032643225514672011

=======================================================
***** Dynamic Programming layer group with cluster ****
=======================================================
total num of base_group is 7
process base group 0, layer_num=1, cluster_num=1
Getting clusters using dynamic programming...
Getting single group cost...
[                       #                       #] 100%
Getting cost_table...
[##################################################] 100%
process base group 1, layer_num=211, cluster_num=125
Searching best group slices...
Getting single group cost...
[##################################################] 100%
Getting cost table...
[##################################################] 100%
process base group 2, layer_num=1, cluster_num=1
process base group 3, layer_num=1, cluster_num=1
process base group 4, layer_num=1, cluster_num=1
process base group 5, layer_num=3, cluster_num=3
[0] [DBG] max_n=4 max_c=64 max_h=256 max_w=16
[0] [DBG] len=320, processed=0, slice=0
[0] [DBG] length split: total=320, processed=0,
              N=1, C=20, H=1, W=16, slice=320
[0] [DBG] len=320, processed=0, slice=320
[0] [DBG] len=320, processed=320, slice=0
[0] [DBG] len=320, processed=320, slice=0
[0] [DBG] max_n=4 max_c=64 max_h=256 max_w=16
[0] [DBG] len=20, processed=0, slice=0
[0] [DBG] length split: total=20, processed=0,
              N=1, C=1, H=1, W=16, slice=16
[0] [DBG] len=20, processed=0, slice=16
[0] [DBG] length split: total=20, processed=16,
              N=1, C=1, H=1, W=4, slice=4
[0] [DBG] len=20, processed=16, slice=4
[0] [DBG] len=20, processed=20, slice=0
[0] [DBG] len=20, processed=20, slice=0
[0] [DBG] max_n=4 max_c=64 max_h=256 max_w=16
[0] [DBG] len=80, processed=0, slice=0
[0] [DBG] length split: total=80, processed=0,
              N=1, C=5, H=1, W=16, slice=80
[0] [DBG] len=80, processed=0, slice=80
[0] [DBG] len=80, processed=80, slice=0
[0] [DBG] len=80, processed=80, slice=0
Searching best group slices...
Getting single group cost...
[                                               #] 100%
Getting cost table...
[                                               #] 100%
process base group 6, layer_num=1, cluster_num=1
================FINAL GROUP================
GroupMethod_process time:15679368
==---------------------------==
Run GroupPostTransformPass : 
    Some transform after layer groups is determined
==---------------------------==
==---------------------------==
Run TimeStepAssignmentPass : 
    Assign timestep task for each group.
==---------------------------==
==---------------------------==
Run LocalMemoryAllocationPass : 
    Allocate local memory for all layer groups
==---------------------------==
==---------------------------==
Run TimeStepCombinePass : 
    Combine time step for better parallel balance
==---------------------------==
===group idx: 1
merge timestep 4 to timestep 3
merge timestep 11 to timestep 10
merge timestep 21 to timestep 20
merge timestep 21 to timestep 20
merge timestep 25 to timestep 24
merge timestep 26 to timestep 25
merge timestep 28 to timestep 27
merge timestep 27 to timestep 26
merge timestep 28 to timestep 27
merge timestep 30 to timestep 29
merge timestep 29 to timestep 28
merge timestep 30 to timestep 29
merge timestep 32 to timestep 31
merge timestep 31 to timestep 30
merge timestep 31 to timestep 30
merge timestep 32 to timestep 31
merge timestep 34 to timestep 33
merge timestep 34 to timestep 33
merge timestep 34 to timestep 33
merge timestep 34 to timestep 33
merge timestep 34 to timestep 33
merge timestep 35 to timestep 34
merge timestep 35 to timestep 34
merge timestep 35 to timestep 34
merge timestep 35 to timestep 34
merge timestep 35 to timestep 34
merge timestep 35 to timestep 34
merge timestep 35 to timestep 34
merge timestep 36 to timestep 35
merge timestep 37 to timestep 36
merge timestep 37 to timestep 36
merge timestep 38 to timestep 37
merge timestep 39 to timestep 38
merge timestep 39 to timestep 38
merge timestep 40 to timestep 39
merge timestep 41 to timestep 40
merge timestep 41 to timestep 40
merge timestep 41 to timestep 40
merge timestep 42 to timestep 41
merge timestep 42 to timestep 41
merge timestep 43 to timestep 42
merge timestep 44 to timestep 43
merge timestep 44 to timestep 43
merge timestep 44 to timestep 43
merge timestep 44 to timestep 43
merge timestep 45 to timestep 44
merge timestep 45 to timestep 44
merge timestep 46 to timestep 45
merge timestep 47 to timestep 46
merge timestep 47 to timestep 46
merge timestep 47 to timestep 46
merge timestep 49 to timestep 48
merge timestep 50 to timestep 49
merge timestep 50 to timestep 49
merge timestep 50 to timestep 49
merge timestep 55 to timestep 54
merge timestep 55 to timestep 54
merge timestep 55 to timestep 54
merge timestep 56 to timestep 55
merge timestep 57 to timestep 56
merge timestep 58 to timestep 57
merge timestep 59 to timestep 58
merge timestep 59 to timestep 58
merge timestep 59 to timestep 58
===group_idx: 2move tensor /model.9/cv2/act/Mul_output_0_Mul from timestep 48 to timestep 0merge timestep 1 to timestep 0
merge timestep 3 to timestep 2
merge timestep 6 to timestep 5
merge timestep 6 to timestep 5
merge timestep 7 to timestep 6
merge timestep 11 to timestep 10
merge timestep 19 to timestep 18
merge timestep 18 to timestep 17
merge timestep 19 to timestep 18
merge timestep 19 to timestep 18
merge timestep 20 to timestep 19
merge timestep 21 to timestep 20
merge timestep 21 to timestep 20
merge timestep 21 to timestep 20
merge timestep 21 to timestep 20
merge timestep 21 to timestep 20
merge timestep 22 to timestep 21
merge timestep 23 to timestep 22
merge timestep 31 to timestep 30
merge timestep 31 to timestep 30
merge timestep 31 to timestep 30
merge timestep 31 to timestep 30
merge timestep 31 to timestep 30
merge timestep 32 to timestep 31
merge timestep 33 to timestep 32
merge timestep 34 to timestep 33
merge timestep 35 to timestep 34
merge timestep 35 to timestep 34
merge timestep 35 to timestep 34
merge timestep 36 to timestep 35
merge timestep 37 to timestep 36
merge timestep 42 to timestep 41
merge timestep 42 to timestep 41
merge timestep 42 to timestep 41
==---------------------------==
Run GroupDataMoveOverlapPass : 
    Overlap data move between two layer group
==---------------------------==
GmemAllocator use FitFirstAssign
[Success]: tpuc-opt yolov8m_bm1684x_f32_tpu.mlir --mlir-disable-threading --strip-io-quant="quant_input=False quant_output=False quant_input_list= quant_output_list= quant_output_bf16=False quant_input_int8=False quant_output_int8=False " --processor-tpu-optimize --dev-parallel --weight-reorder  --subnet-divide="dynamic=False" --op-reorder --topo-sort --layer-group="opt=2 group_by_cores=auto compress_mode=none debugger=0 disable_group_overlap=false lgcache=true config_filename= enable_lghash=False lghash_dir="  --core-parallel  --after-layergroup-weight-reorder --address-assign -o yolov8m_bm1684x_f32_final.mlir --debug_cmd=
[Running]: tpuc-opt yolov8m_bm1684x_f32_final.mlir --codegen="model_file=yolov8m_fp32_1b.bmodel embed_debug_info=False model_version=latest bmodel_only=False gdma_check=True" -o /dev/null
The dir path of compiler_profile is "./"
bmcpu init: skip cpu_user_defined
Cannot open libusercpu.so, disable user cpu layer.
size does not match for tensor /model.0/conv/Conv_output_0_Conv_model.0.conv.weight_swap_channel
readTensor failed
UNREACHABLE executed at /__w/tpu-mlir/tpu-mlir/lib/Support/TensorFile.cpp:293!
PLEASE submit a bug report to https://github.com/llvm/llvm-project/issues/ and include the crash backtrace.
Stack dump:
0.      Program arguments: tpuc-opt yolov8m_bm1684x_f32_final.mlir --init "--codegen=model_file=yolov8m_fp32_1b.bmodel embed_debug_info=False model_version=latest bmodel_only=False gdma_check=True" --deinit --mlir-print-debuginfo -o /dev/null
 #0 0x0000560d9d6fdff7 (/usr/local/lib/python3.10/dist-packages/tpu_mlir/bin/tpuc-opt+0x91cff7)
 #1 0x0000560d9d6fbd1e (/usr/local/lib/python3.10/dist-packages/tpu_mlir/bin/tpuc-opt+0x91ad1e)
 #2 0x0000560d9d6fe97a (/usr/local/lib/python3.10/dist-packages/tpu_mlir/bin/tpuc-opt+0x91d97a)
 #3 0x00007f587b5c1520 (/usr/local/lib/python3.10/dist-packages/tpu_mlir/lib/third_party/libc.so.6+0x42520)
 #4 0x00007f587b6159fc pthread_kill (/usr/local/lib/python3.10/dist-packages/tpu_mlir/lib/third_party/libc.so.6+0x969fc)
 #5 0x00007f587b5c1476 gsignal (/usr/local/lib/python3.10/dist-packages/tpu_mlir/lib/third_party/libc.so.6+0x42476)
 #6 0x00007f587b5a77f3 abort (/usr/local/lib/python3.10/dist-packages/tpu_mlir/lib/third_party/libc.so.6+0x287f3)
 #7 0x0000560d9d6fbb41 (/usr/local/lib/python3.10/dist-packages/tpu_mlir/bin/tpuc-opt+0x91ab41)
 #8 0x0000560d9f1352c2 (/usr/local/lib/python3.10/dist-packages/tpu_mlir/bin/tpuc-opt+0x23542c2)
 #9 0x0000560d9f134dfa (/usr/local/lib/python3.10/dist-packages/tpu_mlir/bin/tpuc-opt+0x2353dfa)
#10 0x0000560d9eeb3c9d (/usr/local/lib/python3.10/dist-packages/tpu_mlir/bin/tpuc-opt+0x20d2c9d)
#11 0x0000560d9eead1f7 (/usr/local/lib/python3.10/dist-packages/tpu_mlir/bin/tpuc-opt+0x20cc1f7)
#12 0x0000560d9e140a4e (/usr/local/lib/python3.10/dist-packages/tpu_mlir/bin/tpuc-opt+0x135fa4e)
#13 0x0000560d9d85034e (/usr/local/lib/python3.10/dist-packages/tpu_mlir/bin/tpuc-opt+0xa6f34e)
#14 0x0000560d9e13ab08 (/usr/local/lib/python3.10/dist-packages/tpu_mlir/bin/tpuc-opt+0x1359b08)
#15 0x0000560d9e138c5b (/usr/local/lib/python3.10/dist-packages/tpu_mlir/bin/tpuc-opt+0x1357c5b)
#16 0x0000560d9e13593e (/usr/local/lib/python3.10/dist-packages/tpu_mlir/bin/tpuc-opt+0x135493e)
#17 0x0000560d9f008b04 (/usr/local/lib/python3.10/dist-packages/tpu_mlir/bin/tpuc-opt+0x2227b04)
#18 0x0000560d9f009131 (/usr/local/lib/python3.10/dist-packages/tpu_mlir/bin/tpuc-opt+0x2228131)
#19 0x0000560d9f00b5d8 (/usr/local/lib/python3.10/dist-packages/tpu_mlir/bin/tpuc-opt+0x222a5d8)
#20 0x0000560d9d6ef67b (/usr/local/lib/python3.10/dist-packages/tpu_mlir/bin/tpuc-opt+0x90e67b)
#21 0x0000560d9d6eea44 (/usr/local/lib/python3.10/dist-packages/tpu_mlir/bin/tpuc-opt+0x90da44)
#22 0x0000560d9f226058 (/usr/local/lib/python3.10/dist-packages/tpu_mlir/bin/tpuc-opt+0x2445058)
#23 0x0000560d9d6e8d4a (/usr/local/lib/python3.10/dist-packages/tpu_mlir/bin/tpuc-opt+0x907d4a)
#24 0x0000560d9d6e9214 (/usr/local/lib/python3.10/dist-packages/tpu_mlir/bin/tpuc-opt+0x908214)
#25 0x0000560d9d6e7c5a (/usr/local/lib/python3.10/dist-packages/tpu_mlir/bin/tpuc-opt+0x906c5a)
#26 0x00007f587b5a8d90 (/usr/local/lib/python3.10/dist-packages/tpu_mlir/lib/third_party/libc.so.6+0x29d90)
#27 0x00007f587b5a8e40 __libc_start_main (/usr/local/lib/python3.10/dist-packages/tpu_mlir/lib/third_party/libc.so.6+0x29e40)
#28 0x0000560d9d6e7065 (/usr/local/lib/python3.10/dist-packages/tpu_mlir/bin/tpuc-opt+0x906065)
Aborted (core dumped)
Traceback (most recent call last):
  File "/usr/local/lib/python3.10/dist-packages/tpu_mlir/python/tools/model_deploy.py", line 678, in <module>
    tpu_patterns = tool.build_model()
  File "/usr/local/lib/python3.10/dist-packages/tpu_mlir/python/tools/model_deploy.py", line 402, in build_model
    patterns = mlir_to_model(
  File "/usr/local/lib/python3.10/dist-packages/tpu_mlir/python/utils/mlir_shell.py", line 1066, in mlir_to_model
    _os_system(cmd, log_level=log_level)
  File "/usr/local/lib/python3.10/dist-packages/tpu_mlir/python/utils/mlir_shell.py", line 495, in _os_system
    raise RuntimeError("[!Error]: {}".format(cmd_str))
RuntimeError: [!Error]: tpuc-opt yolov8m_bm1684x_f32_final.mlir --codegen="model_file=yolov8m_fp32_1b.bmodel embed_debug_info=False model_version=latest bmodel_only=False gdma_check=True" -o /dev/null
mv: cannot stat 'yolov8m_fp32_1b.bmodel': No such file or directory