Skip to content

Bug: w4bf16 quantization on Qwen3.5-2B fails with "invalid scale channel" in A16MatMul.cpp for BM1684X on tpu_mlir-1.28.1 #269

Open
Open
Bug: w4bf16 quantization on Qwen3.5-2B fails with "invalid scale channel" in A16MatMul.cpp for BM1684X on tpu_mlir-1.28.1#269
Labels
bugSomething isn't working

Description

@moonnryan
moonnryan
opened on May 14, 2026

Describe the bug

When converting Qwen3.5-2B model with w4bf16 quantization for BM1684X, tpuc-opt crashes with the error:
invalid scale channel UNREACHABLE executed at A16MatMul.cpp:274

The root cause is a hard check in A16MatMul.cpp that requires the scale channel count to be divisible by NPU_NUM (64) for BM1684X.
For Qwen3.5-2B with w4bf16 quantization (both q_group_size=128 and 64 tested), the resulting scale shape does not satisfy scale_shape[0] % 64 == 0, leading to an unreachable assertion failure.

This blocks 4-bit weight-only quantization support for Qwen3.5-2B on BM1684X, while bf16 full-precision conversion works fine.

Steps to reproduce

1. Environment: tpu-mlir latest docker, BM1684X, Qwen3.5-2B (HuggingFace format)
2. Run the conversion command:
   
   llm_convert.py -m /workspace/Qwen3.5-2B \
   --max_input_length 6144 \
   -s 6144 \
   --quantize w4bf16 \
   --q_group_size 64 \
   -c bm1684x \
   --out_dir qwen3.5_2b
3. Observe tpuc-opt crash during parallel compilation of transformer blocks with the error:
4. invalid scale channel UNREACHABLE executed at /__w/tpu-mlir/tpu-mlir/lib/Dialect/Tpu/Transforms/WeightReorder/BM1684X/A16MatMul.cpp:274!
This happens on multiple blocks (e.g., block_2, block_8, block_18, block_21) regardless of q_group_size=64 or 128.

Expected behavior

Qwen3.5-2B with w4bf16 quantization should compile without assertion failure, as supported by previous models (Qwen2, Llama) on BM1684X.

Error logs / stack trace

The crash comes from this hard check in model/transforms/weight_reorder/BM1684X/A16MatMul.cpp:
int64_t npu_num = backend::Arch::NPU_NUM;
if (scale_shape[0] % npu_num) {
  llvm_unreachable("invalid scale channel");
}
fail log:
llm_convert.py -m /workspace/Qwen3.5-2B-int4-AutoRound --max_input_length 1024  -s 2048  -c bm1684x --out_dir qwen3.5  --max_pixels 768,768
generate vit  mlir ...
generate embedding and lm_head mlir ...
generate block_0 linear attention mlir ...
generate block_1 linear attention mlir ...
generate block_2 linear attention mlir ...
generate block_3 full attention mlir ...
generate block_4 linear attention mlir ...
generate block_5 linear attention mlir ...
generate block_6 linear attention mlir ...
generate block_7 full attention mlir ...
generate block_8 linear attention mlir ...
generate block_9 linear attention mlir ...
generate block_10 linear attention mlir ...
generate block_11 full attention mlir ...
generate block_12 linear attention mlir ...
generate block_13 linear attention mlir ...
generate block_14 linear attention mlir ...
generate block_15 full attention mlir ...
generate block_16 linear attention mlir ...
generate block_17 linear attention mlir ...
generate block_18 linear attention mlir ...
generate block_19 full attention mlir ...
generate block_20 linear attention mlir ...
generate block_21 linear attention mlir ...
generate block_22 linear attention mlir ...
generate block_23 full attention mlir ...
generate mlir: 100%|██████████████████████████████████████████████████████████████████████████████████████████████████████████████| 26/26 [00:24<00:00,  1.07it/s]
Executing command:
bash -c parallel -j 48 --halt now,fail=1 --progress --joblog task.txt.log < task.txt

Computers / CPU cores / Max jobs to run
1:local / 48 / 48

Computer:jobs running/jobs completed/%of started jobs/Average seconds to complete
local:48/0/100%/0.0s invalid scale channel
UNREACHABLE executed at /__w/tpu-mlir/tpu-mlir/lib/Dialect/Tpu/Transforms/WeightReorder/BM1684X/A16MatMul.cpp:274!
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 block_18_bm1684x_w4bf16_tpu.mlir --init --mlir-disable-threading "--strip-io-quant=quant_input=True quant_output=True 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=True --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 --deinit --mlir-print-debuginfo -o block_18_bm1684x_w4bf16_final.mlir
 #0 0x000055db992016e7 (/usr/local/lib/python3.10/dist-packages/tpu_mlir/bin/tpuc-opt+0x92d6e7)
 #1 0x000055db991ff40e (/usr/local/lib/python3.10/dist-packages/tpu_mlir/bin/tpuc-opt+0x92b40e)
 #2 0x000055db9920206a (/usr/local/lib/python3.10/dist-packages/tpu_mlir/bin/tpuc-opt+0x92e06a)
 #3 0x00007f50090e8520 (/usr/local/lib/python3.10/dist-packages/tpu_mlir/lib/third_party/libc.so.6+0x42520)
 #4 0x00007f500913c9fc pthread_kill (/usr/local/lib/python3.10/dist-packages/tpu_mlir/lib/third_party/libc.so.6+0x969fc)
 #5 0x00007f50090e8476 gsignal (/usr/local/lib/python3.10/dist-packages/tpu_mlir/lib/third_party/libc.so.6+0x42476)
 #6 0x00007f50090ce7f3 abort (/usr/local/lib/python3.10/dist-packages/tpu_mlir/lib/third_party/libc.so.6+0x287f3)
 #7 0x000055db991ff231 (/usr/local/lib/python3.10/dist-packages/tpu_mlir/bin/tpuc-opt+0x92b231)
 #8 0x000055db9a018e40 (/usr/local/lib/python3.10/dist-packages/tpu_mlir/bin/tpuc-opt+0x1744e40)
 #9 0x000055db9a018e98 (/usr/local/lib/python3.10/dist-packages/tpu_mlir/bin/tpuc-opt+0x1744e98)
#10 0x000055db99ffde14 (/usr/local/lib/python3.10/dist-packages/tpu_mlir/bin/tpuc-opt+0x1729e14)
#11 0x000055db9ab873f7 (/usr/local/lib/python3.10/dist-packages/tpu_mlir/bin/tpuc-opt+0x22b33f7)
#12 0x000055db9ab83d0f (/usr/local/lib/python3.10/dist-packages/tpu_mlir/bin/tpuc-opt+0x22afd0f)
#13 0x000055db9ab4cf0c (/usr/local/lib/python3.10/dist-packages/tpu_mlir/bin/tpuc-opt+0x2278f0c)
#14 0x000055db9ab49d5c (/usr/local/lib/python3.10/dist-packages/tpu_mlir/bin/tpuc-opt+0x2275d5c)
#15 0x000055db99fc559e (/usr/local/lib/python3.10/dist-packages/tpu_mlir/bin/tpuc-opt+0x16f159e)
#16 0x000055db9abb06e4 (/usr/local/lib/python3.10/dist-packages/tpu_mlir/bin/tpuc-opt+0x22dc6e4)
#17 0x000055db9abb0d11 (/usr/local/lib/python3.10/dist-packages/tpu_mlir/bin/tpuc-opt+0x22dcd11)
#18 0x000055db9abb31b8 (/usr/local/lib/python3.10/dist-packages/tpu_mlir/bin/tpuc-opt+0x22df1b8)
#19 0x000055db991f2d6b (/usr/local/lib/python3.10/dist-packages/tpu_mlir/bin/tpuc-opt+0x91ed6b)
#20 0x000055db991f2134 (/usr/local/lib/python3.10/dist-packages/tpu_mlir/bin/tpuc-opt+0x91e134)
#21 0x000055db9adce308 (/usr/local/lib/python3.10/dist-packages/tpu_mlir/bin/tpuc-opt+0x24fa308)
#22 0x000055db991ec43a (/usr/local/lib/python3.10/dist-packages/tpu_mlir/bin/tpuc-opt+0x91843a)
#23 0x000055db991ec904 (/usr/local/lib/python3.10/dist-packages/tpu_mlir/bin/tpuc-opt+0x918904)
#24 0x000055db991eb34a (/usr/local/lib/python3.10/dist-packages/tpu_mlir/bin/tpuc-opt+0x91734a)
#25 0x00007f50090cfd90 (/usr/local/lib/python3.10/dist-packages/tpu_mlir/lib/third_party/libc.so.6+0x29d90)
#26 0x00007f50090cfe40 __libc_start_main (/usr/local/lib/python3.10/dist-packages/tpu_mlir/lib/third_party/libc.so.6+0x29e40)
#27 0x000055db991ea755 (/usr/local/lib/python3.10/dist-packages/tpu_mlir/bin/tpuc-opt+0x916755)
Aborted
Traceback (most recent call last):
  File "/usr/local/lib/python3.10/dist-packages/tpu_mlir/python/tools/model_deploy.py", line 698, in <module>
    tpu_patterns = tool.build_model()
  File "/usr/local/lib/python3.10/dist-packages/tpu_mlir/python/tools/model_deploy.py", line 410, in build_model
    patterns = mlir_to_model(
  File "/usr/local/lib/python3.10/dist-packages/tpu_mlir/python/utils/mlir_shell.py", line 1058, 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 block_18_bm1684x_w4bf16_tpu.mlir --mlir-disable-threading --strip-io-quant="quant_input=True quant_output=True 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=True" --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 block_18_bm1684x_w4bf16_final.mlir --debug_cmd=
parallel: This job failed:
(pushd block_18 &&  model_deploy.py --mlir block_18.mlir --quantize w4bf16 --q_group_size 128 --quant_input --quant_output --chip bm1684x --num_core 1 --num_device 1 --model block_18.bmodel --addr_mode basic --high_precision --dynamic && popd ) > block_18.log
invalid scale channel
UNREACHABLE executed at /__w/tpu-mlir/tpu-mlir/lib/Dialect/Tpu/Transforms/WeightReorder/BM1684X/A16MatMul.cpp:274!
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 block_17_bm1684x_w4bf16_tpu.mlir --init --mlir-disable-threading "--strip-io-quant=quant_input=True quant_output=True 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=True --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 --deinit --mlir-print-debuginfo -o block_17_bm1684x_w4bf16_final.mlir
 #0 0x000055ec6e9d46e7 (/usr/local/lib/python3.10/dist-packages/tpu_mlir/bin/tpuc-opt+0x92d6e7)
 #1 0x000055ec6e9d240e (/usr/local/lib/python3.10/dist-packages/tpu_mlir/bin/tpuc-opt+0x92b40e)
 #2 0x000055ec6e9d506a (/usr/local/lib/python3.10/dist-packages/tpu_mlir/bin/tpuc-opt+0x92e06a)
 #3 0x00007fa379f7a520 (/usr/local/lib/python3.10/dist-packages/tpu_mlir/lib/third_party/libc.so.6+0x42520)
 #4 0x00007fa379fce9fc pthread_kill (/usr/local/lib/python3.10/dist-packages/tpu_mlir/lib/third_party/libc.so.6+0x969fc)
 #5 0x00007fa379f7a476 gsignal (/usr/local/lib/python3.10/dist-packages/tpu_mlir/lib/third_party/libc.so.6+0x42476)
 #6 0x00007fa379f607f3 abort (/usr/local/lib/python3.10/dist-packages/tpu_mlir/lib/third_party/libc.so.6+0x287f3)
 #7 0x000055ec6e9d2231 (/usr/local/lib/python3.10/dist-packages/tpu_mlir/bin/tpuc-opt+0x92b231)
 #8 0x000055ec6f7ebe40 (/usr/local/lib/python3.10/dist-packages/tpu_mlir/bin/tpuc-opt+0x1744e40)
 #9 0x000055ec6f7ebe98 (/usr/local/lib/python3.10/dist-packages/tpu_mlir/bin/tpuc-opt+0x1744e98)
#10 0x000055ec6f7d0e14 (/usr/local/lib/python3.10/dist-packages/tpu_mlir/bin/tpuc-opt+0x1729e14)
#11 0x000055ec7035a3f7 (/usr/local/lib/python3.10/dist-packages/tpu_mlir/bin/tpuc-opt+0x22b33f7)
#12 0x000055ec70356d0f (/usr/local/lib/python3.10/dist-packages/tpu_mlir/bin/tpuc-opt+0x22afd0f)
#13 0x000055ec7031ff0c (/usr/local/lib/python3.10/dist-packages/tpu_mlir/bin/tpuc-opt+0x2278f0c)
#14 0x000055ec7031cd5c (/usr/local/lib/python3.10/dist-packages/tpu_mlir/bin/tpuc-opt+0x2275d5c)
#15 0x000055ec6f79859e (/usr/local/lib/python3.10/dist-packages/tpu_mlir/bin/tpuc-opt+0x16f159e)
#16 0x000055ec703836e4 (/usr/local/lib/python3.10/dist-packages/tpu_mlir/bin/tpuc-opt+0x22dc6e4)
#17 0x000055ec70383d11 (/usr/local/lib/python3.10/dist-packages/tpu_mlir/bin/tpuc-opt+0x22dcd11)
#18 0x000055ec703861b8 (/usr/local/lib/python3.10/dist-packages/tpu_mlir/bin/tpuc-opt+0x22df1b8)
#19 0x000055ec6e9c5d6b (/usr/local/lib/python3.10/dist-packages/tpu_mlir/bin/tpuc-opt+0x91ed6b)
#20 0x000055ec6e9c5134 (/usr/local/lib/python3.10/dist-packages/tpu_mlir/bin/tpuc-opt+0x91e134)
#21 0x000055ec705a1308 (/usr/local/lib/python3.10/dist-packages/tpu_mlir/bin/tpuc-opt+0x24fa308)
#22 0x000055ec6e9bf43a (/usr/local/lib/python3.10/dist-packages/tpu_mlir/bin/tpuc-opt+0x91843a)
#23 0x000055ec6e9bf904 (/usr/local/lib/python3.10/dist-packages/tpu_mlir/bin/tpuc-opt+0x918904)
#24 0x000055ec6e9be34a (/usr/local/lib/python3.10/dist-packages/tpu_mlir/bin/tpuc-opt+0x91734a)
#25 0x00007fa379f61d90 (/usr/local/lib/python3.10/dist-packages/tpu_mlir/lib/third_party/libc.so.6+0x29d90)
#26 0x00007fa379f61e40 __libc_start_main (/usr/local/lib/python3.10/dist-packages/tpu_mlir/lib/third_party/libc.so.6+0x29e40)
#27 0x000055ec6e9bd755 (/usr/local/lib/python3.10/dist-packages/tpu_mlir/bin/tpuc-opt+0x916755)
Aborted
Traceback (most recent call last):
  File "/usr/local/lib/python3.10/dist-packages/tpu_mlir/python/tools/model_deploy.py", line 698, in <module>
    tpu_patterns = tool.build_model()
  File "/usr/local/lib/python3.10/dist-packages/tpu_mlir/python/tools/model_deploy.py", line 410, in build_model
    patterns = mlir_to_model(
  File "/usr/local/lib/python3.10/dist-packages/tpu_mlir/python/utils/mlir_shell.py", line 1058, 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 block_17_bm1684x_w4bf16_tpu.mlir --mlir-disable-threading --strip-io-quant="quant_input=True quant_output=True 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=True" --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 block_17_bm1684x_w4bf16_final.mlir --debug_cmd=
parallel: This job failed:
(pushd block_17 &&  model_deploy.py --mlir block_17.mlir --quantize w4bf16 --q_group_size 128 --quant_input --quant_output --chip bm1684x --num_core 1 --num_device 1 --model block_17.bmodel --addr_mode basic --high_precision --dynamic && popd ) > block_17.log
invalid scale channel
UNREACHABLE executed at /__w/tpu-mlir/tpu-mlir/lib/Dialect/Tpu/Transforms/WeightReorder/BM1684X/A16MatMul.cpp:274!
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 block_cache_20_bm1684x_w4bf16_tpu.mlir --init --mlir-disable-threading "--strip-io-quant=quant_input=True quant_output=True 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=same_addr=0:0,1:1,2:2 --deinit --mlir-print-debuginfo -o block_cache_20_bm1684x_w4bf16_final.mlir
 #0 0x000055a8030186e7 (/usr/local/lib/python3.10/dist-packages/tpu_mlir/bin/tpuc-opt+0x92d6e7)
 #1 0x000055a80301640e (/usr/local/lib/python3.10/dist-packages/tpu_mlir/bin/tpuc-opt+0x92b40e)
 #2 0x000055a80301906a (/usr/local/lib/python3.10/dist-packages/tpu_mlir/bin/tpuc-opt+0x92e06a)
 #3 0x00007f3abd102520 (/usr/local/lib/python3.10/dist-packages/tpu_mlir/lib/third_party/libc.so.6+0x42520)
 #4 0x00007f3abd1569fc pthread_kill (/usr/local/lib/python3.10/dist-packages/tpu_mlir/lib/third_party/libc.so.6+0x969fc)
 #5 0x00007f3abd102476 gsignal (/usr/local/lib/python3.10/dist-packages/tpu_mlir/lib/third_party/libc.so.6+0x42476)
 #6 0x00007f3abd0e87f3 abort (/usr/local/lib/python3.10/dist-packages/tpu_mlir/lib/third_party/libc.so.6+0x287f3)
 #7 0x000055a803016231 (/usr/local/lib/python3.10/dist-packages/tpu_mlir/bin/tpuc-opt+0x92b231)
 #8 0x000055a803e2fe40 (/usr/local/lib/python3.10/dist-packages/tpu_mlir/bin/tpuc-opt+0x1744e40)
 #9 0x000055a803e2fe98 (/usr/local/lib/python3.10/dist-packages/tpu_mlir/bin/tpuc-opt+0x1744e98)
#10 0x000055a803e14e14 (/usr/local/lib/python3.10/dist-packages/tpu_mlir/bin/tpuc-opt+0x1729e14)
#11 0x000055a80499e3f7 (/usr/local/lib/python3.10/dist-packages/tpu_mlir/bin/tpuc-opt+0x22b33f7)
#12 0x000055a80499ad0f (/usr/local/lib/python3.10/dist-packages/tpu_mlir/bin/tpuc-opt+0x22afd0f)
#13 0x000055a804963f0c (/usr/local/lib/python3.10/dist-packages/tpu_mlir/bin/tpuc-opt+0x2278f0c)
#14 0x000055a804960d5c (/usr/local/lib/python3.10/dist-packages/tpu_mlir/bin/tpuc-opt+0x2275d5c)
#15 0x000055a803ddc59e (/usr/local/lib/python3.10/dist-packages/tpu_mlir/bin/tpuc-opt+0x16f159e)
#16 0x000055a8049c76e4 (/usr/local/lib/python3.10/dist-packages/tpu_mlir/bin/tpuc-opt+0x22dc6e4)
#17 0x000055a8049c7d11 (/usr/local/lib/python3.10/dist-packages/tpu_mlir/bin/tpuc-opt+0x22dcd11)
#18 0x000055a8049ca1b8 (/usr/local/lib/python3.10/dist-packages/tpu_mlir/bin/tpuc-opt+0x22df1b8)
#19 0x000055a803009d6b (/usr/local/lib/python3.10/dist-packages/tpu_mlir/bin/tpuc-opt+0x91ed6b)
#20 0x000055a803009134 (/usr/local/lib/python3.10/dist-packages/tpu_mlir/bin/tpuc-opt+0x91e134)
#21 0x000055a804be5308 (/usr/local/lib/python3.10/dist-packages/tpu_mlir/bin/tpuc-opt+0x24fa308)
#22 0x000055a80300343a (/usr/local/lib/python3.10/dist-packages/tpu_mlir/bin/tpuc-opt+0x91843a)
#23 0x000055a803003904 (/usr/local/lib/python3.10/dist-packages/tpu_mlir/bin/tpuc-opt+0x918904)
#24 0x000055a80300234a (/usr/local/lib/python3.10/dist-packages/tpu_mlir/bin/tpuc-opt+0x91734a)
#25 0x00007f3abd0e9d90 (/usr/local/lib/python3.10/dist-packages/tpu_mlir/lib/third_party/libc.so.6+0x29d90)
#26 0x00007f3abd0e9e40 __libc_start_main (/usr/local/lib/python3.10/dist-packages/tpu_mlir/lib/third_party/libc.so.6+0x29e40)
#27 0x000055a803001755 (/usr/local/lib/python3.10/dist-packages/tpu_mlir/bin/tpuc-opt+0x916755)
Aborted
Traceback (most recent call last):
  File "/usr/local/lib/python3.10/dist-packages/tpu_mlir/python/tools/model_deploy.py", line 698, in <module>
    tpu_patterns = tool.build_model()
  File "/usr/local/lib/python3.10/dist-packages/tpu_mlir/python/tools/model_deploy.py", line 410, in build_model
    patterns = mlir_to_model(
  File "/usr/local/lib/python3.10/dist-packages/tpu_mlir/python/utils/mlir_shell.py", line 1058, 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 block_cache_20_bm1684x_w4bf16_tpu.mlir --mlir-disable-threading --strip-io-quant="quant_input=True quant_output=True 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="same_addr=0:0,1:1,2:2" -o block_cache_20_bm1684x_w4bf16_final.mlir --debug_cmd=
parallel: This job failed:
(pushd block_cache_20 &&  model_deploy.py --mlir block_cache_20.mlir --quantize w4bf16 --q_group_size 128 --quant_input --quant_output --chip bm1684x --addr_mode io_alone --num_core 1 --num_device 1 --model block_cache_20.bmodel --high_precision --same_addr 0:0,1:1,2:2 && popd ) > block_cache_20.log
Error: Command failed with return code 1
Failed command: bash -c parallel -j 48 --halt now,fail=1 --progress --joblog task.txt.log < task.txt

TPU-MLIR version

tpu_mlir-1.28.1

Target chip

bm1684x

Source framework / model

Qwen/Qwen3.5-2B && Intel/Qwen3.5-2B-int4-AutoRound

OS / Docker image

sophgo/tpuc_dev:latest

Metadata

Metadata

Assignees

No one assigned

    Labels

    bugSomething isn't working

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions