자체 텐서플로우 모델을 변환하여 furiosa litmus로 테스트하니 아래와 같은 오류가 나는데요. 해결하려면 어떻게 접근해야 할까요?
libfuriosa_hal.so --- v0.11.0, built @ 43c901f
INFO:furiosa.common.native:loaded native library libfuriosa_compiler.so.0.10.0 (0.10.0 f8f05c8ea)
libfuriosa_hal.so --- v0.11.0, built @ 43c901f
furiosa-quantizer 0.10.0 (rev. ded42da) furiosa-litmus 0.10.0 (rev. ded42da)
[Step 1] Checking if the model can be loaded and optimized ...
INFO:Furiosa-Optimizer:Dynamic batch size is detected at input_name: rescaling_input. Fix batch_size=1 for valid shape inference.
[Step 1] Passed
[Step 2] Checking if the model can be quantized ...
[Step 2] Passed
[Step 3] Checking if the model can be compiled for the NPU family [warboy-2pe] ...
[1/6] 🔍 Compiling from onnx to dfg
Done in 1.6340518s
[2/6] 🔍 Compiling from dfg to ldfg
▪▪▪▪▪ [1/3] Splitting graph(LAS)...Done in 4.3336773s
ERROR: Failed to split operator#7: Conv2d: operator cannot be split under the memory budget:
Conv2d
Padding { top: 0, left: 0, bottom: 0, right: 0 }
stride: (H: 1, W: 1), dilation: (H: 1, W: 1),
clipping: (Min: 0, Max: inf),
groups: 1
name:
input tensors: 3
input tensor 71: [1x1x1x320000] NxHxWxC, 320000 B, i8
(axes: [], <(m:0, M:0.10867147147655487) x 1>)
(s: 0.00042616263324139167, o: -128) x 1
source: none
input tensor 86: [256x1x1x320000] NxHxWxC, 81920000 B, i8
(axes: [AxisIndex(0)], <(m:-0.004732204234506201, M:0.004695233888924122) x 256>)
(s: 0.0000369703455820797, o: 0) x 256
source: none
input tensor 88: [256] W, 1024 B, i32
(axes: [AxisIndex(0)], <(m:-33.83442054243851, M:33.834420526683125) x 256>)
(s: 0.000000015755379825103333, o: 0) x 256
source: unknown
total bytes: 82241024
output tensors: 1
output tensor 103: [1x1x1x256] NxHxWxC, 256 B, i8
(axes: [], <(m:0, M:2.014754295349121) x 1>)
(s: 0.00790099723666322, o: -128) x 1
source: unknown
total bytes: 256
[Step 3] Failed
Traceback (most recent call last):
File "/home/kykang/.local/share/virtualenvs/furiosa-LqRhWH5A/bin/furiosa-litmus", line 8, in <module>
sys.exit(main())
File "/home/kykang/.local/share/virtualenvs/furiosa-LqRhWH5A/lib/python3.8/site-packages/furiosa/litmus/__init__.py", line 254, in main
validate(
File "/home/kykang/.local/share/virtualenvs/furiosa-LqRhWH5A/lib/python3.8/site-packages/furiosa/litmus/__init__.py", line 219, in validate
enf_path = step3_compile(quantized_model, reporter, target_npu, verbose, tmpdir)
File "/home/kykang/.local/share/virtualenvs/furiosa-LqRhWH5A/lib/python3.8/site-packages/furiosa/litmus/__init__.py", line 160, in step3_compile
raise e
File "/home/kykang/.local/share/virtualenvs/furiosa-LqRhWH5A/lib/python3.8/site-packages/furiosa/litmus/__init__.py", line 145, in step3_compile
enf = compile(
File "/home/kykang/.local/share/virtualenvs/furiosa-LqRhWH5A/lib/python3.8/site-packages/furiosa/tools/compiler/api/v1.py", line 226, in compile
raise into_exception(errno)
furiosa.tools.compiler.api.errors.InternalError: Unknown (native error code: 255)