pranav-prakash
commented
4 years ago Hm I've never seen that "Error message: Protobuf parsing failed" before. That syscall seems to correspond to tgkill so I think that's just the runner trying to kill itself after the failed protobuf parse.
I cloned a fresh version of the repo (on centos7) and couldn't reproduce the issue, which is going to make this a bit harder to debug. (Just as a quick sanity-check, can you run md5sum googlenet_quantized.onnx? It should be e1360471e07e0810ee3696eb44e66c57).
The only significant difference I can see at the moment is that your version of pk / spike seems to be a bit newer than what I've been using (in particular mine doesn't print "bbl loader").
Can you try building spike via the following:
-
Clone https://github.com/ucb-bar/esp-tools (using the default master branch, latest commit should be
dcb6012f7) -
Apply the syscall patches inside the riscv-pk folder
-
Inside riscv-isa-sim, checkout commit
506d83f6a8don the master branch -
Build spike and try using that built version of spike to run.
Hopefully this works, if not we can do some more digging (the fact that protobuf parsing fails seems to indicate that it's a pretty fundamental issue unrelated to the gemmini extension).
As for the calibration script, it's written to assume that the --dataset_path directory contains a bunch of folders labeled test_data_set_1, test_data_set_2, etc. each of which contains an input_0.pb, input_1.pb, and so on. I haven't checked what the folder structure of the arcface model looks like, but the error seems to indicate that the folder structure might not match what's described above.
Also do note that the imagenet_runner was hard-coded for using fixed 224x224 pngs as input, so you'll probably have to create your own runner for arcface, especially since it seems that the model can accept variable-sized images (so you'll have to manually set the input tensor dimensions – you can look at the upstream onnxruntime api docs for more info).
Also last point for the mxnet derived models is that I've found them to be very finnicky and sensitive to quantization. You might want to first get the non-quantized (i.e. floating point) model running first (for which you'll likely need to write your own runner), at which point you can start playing around with quantization. The accuracy for ResNet models is also a bit poor at the moment because we're limited to power-of-2 scale factors. I've also never really tried the quantization/calibration scripts with non-imagenet models (although they should presumably work since they were modified from microsoft's upstream implementation), so I'd be interested to hear how it turns out!
J-Zenk
Describe the bug When run onnx models here
https://github.com/pranav-prakash/onnxruntime-riscv/releases/tag/v0.01, I gotbad syscall #131!I have tried googlenet_quantized.onnx, mobilenet_quantized_optimized.onnx and resnet50_quantized.onnx, only the resnet could run normally.System information
To Reproduce
https://github.com/pranav-prakash/onnxruntime-riscv/releases/tag/v0.01spike --extension=gemmini pk ort_test -m googlenet_quantized.onnx -i images/cat.jpg -p caffe2 -x 1 -O 0Here is the resultbad syscall #131!
Traceback (most recent call last): File "calibrate.py", line 379, in
main()
File "calibrate.py", line 348, in main
args.data_preprocess)
File "calibrate.py", line 289, in load_test_data
preprocess_method)
File "calibrate.py", line 261, in load_single_test_data
'Number of input protobufs does not match expected model inputs')
ValueError: Number of input protobufs does not match expected model inputs