Frank comments

Results 11 comments of


                                            Frank

Export ONNX model error

> > torch: 1.12.0+cu116 > > hello, have you solved the problem ?i meet the same problem when i run pytorch2onnx.py No. I guess this is a problem of MMSeg...

I change the signature and body of the forward function as following: ``` @auto_fp16(apply_to=('img', )) def forward(self, img, **kwargs): # img_metas, return_loss=True """Calls either :func:`forward_train` or :func:`forward_test` depending on whether...

AssertionError: ADE20KDataset: MultiScaleFlipAug:

From your error trace, I see `assert mmcv.is_list_of(img_ratios, float)`. Maybe the `img_ratios` is not a list of floating numbers. Check it or remove `img_ratios` from your config file.

Possibility of adding a working demo code

I don't think one is willing to use slim to try MorphNet...

Possibility of adding a working demo code

@eladeban Considering Keras is so popular and has a close relation with TensorFlow, so I think a model in Keras is a good start. What do you think ResNet50?

Why ORT_lightning_X_trt_complete didn't compare ONNX Runtime and PyTorch results?

I have the same problem: ``` AssertionError: Not equal to tolerance rtol=0.001, atol=1e-05 Mismatched elements: 4 / 4 (100%) Max absolute difference: 0.11626142 Max relative difference: 0.2094227 x: array([[0.477019, 0.421438,...

stark_lighting model convert onnx error

The exported pytorch models are both ok.

llama runner process no longer running: -1 CUDA error: CUBLAS_STATUS_EXECUTION_FAILED

same error on AGX Orin 64GB, Jetpack 6.0 DP [L4T 36.2.0]