aimet_encoding error

#2
by chencong915 - opened

Hi, I followed the instructions to run the yolov8_det_quantized model demo and got following error. Can you please help me fix the issue. Thanks.

Traceback (most recent call last):
File "***********************lib/python3.8/runpy.py", line 194, in _run_module_as_main
return _run_code(code, main_globals, None,
File "***********************lib/python3.8/runpy.py", line 87, in _run_code
exec(code, run_globals)
File "***********************lib/python3.8/site-packages/qai_hub_models/models/yolov8_det_quantized/demo.py", line 25, in
main()
File "***********************lib/python3.8/site-packages/qai_hub_models/models/yolov8_det_quantized/demo.py", line 15, in main
yolo_detection_demo(
File "***********************lib/python3.8/site-packages/qai_hub_models/models/_shared/yolo/demo.py", line 57, in yolo_detection_demo
model = demo_model_from_cli_args(model_type, model_id, args)
File "***********************lib/python3.8/site-packages/qai_hub_models/utils/args.py", line 323, in demo_model_from_cli_args
inference_model = model_from_cli_args(model_cls, cli_args)
File "***********************lib/python3.8/site-packages/qai_hub_models/utils/args.py", line 262, in model_from_cli_args
return model_cls.from_pretrained(**get_model_kwargs(model_cls, vars(cli_args)))
File "***********************lib/python3.8/site-packages/qai_hub_models/models/yolov8_det_quantized/model.py", line 85, in from_pretrained
load_encodings_to_sim(sim, aimet_encodings)
File "***********************lib/python3.8/site-packages/aimet_torch/utils.py", line 1205, in fn_wrapper
return _callable(*args, **kwargs)
File "***********************lib/python3.8/site-packages/aimet_torch/quantsim.py", line 2195, in load_encodings_to_sim
quant_sim_model.load_encodings(pytorch_encoding_path,
File "***********************lib/python3.8/site-packages/aimet_torch/quantsim.py", line 1687, in load_encodings
self._load_encodings_impl(encodings, strict, partial, requires_grad, allow_overwrite)
File "***********************lib/python3.8/site-packages/aimet_torch/quantsim.py", line 1715, in _load_encodings_impl
raise RuntimeError(msg)
RuntimeError: Encoding dictionary contains modules/parameters that doesn't exist in the model: model.model.22.module_truediv, module_truediv_2

Qualcomm org

Thanks for reporting. Will look into it.

It works for me now by changing the model loading from
.from_pretrained(
ckpt_name="yolov8s.pt"
)

to

.from_pretrained(
ckpt_name="yolov8s.pt", aimet_encodings=None, include_postprocessing=True
)

Qualcomm org
edited Oct 11

This is fixed in our latest release, v0.15.0. Let us know if you still hit the issue.

korywat changed discussion status to closed

Sign up or log in to comment