torch2trt

torch2trt copied to clipboard

Hi, - with `tensorrt_converter` for instance norm 3D, the result was abnormal - Although verification by `pytest` passes, instance norm with the default initialization. random initialization will be failed. https://github.com/NVIDIA-AI-IOT/torch2trt/blob/master/torch2trt/converters/native_converters.py?plain=1#L902...

Yuxiang1990

When I run the instruction: sudo python3 setup.py install --plugins I get an error: Illegal instruction.

CMeiyi

AttributeError when using torch.nn.functional.pad

3

Hi all, when I run conversion of the following operation: ```python x_padded = torch.nn.functional.pad(x, (0, 0, pad_left, pad_right)) ``` I get the error below: ```bash AttributeError: 'tensorrt_bindings.tensorrt.INetworkDefinition' object has no...

clementabary

taotaoorange

Is that available on orin nano?

2

I have a jetson Orin Nano model. I've tried to install torch2trt in my jetson device using docker system. Here is my error log during installation. root@ubuntu:/home/torch2trt# python3 setup.py install...

adrenalin7237

Because how flattener limits to floats as inputs, it will wrongly remove any int inputs that are common in text models e.g. encoded token ids. Discovered when trying to convert...

aliencaocao

``` import torch from transformers import AutoImageProcessor, AutoModelForZeroShotImageClassification, AutoTokenizer, ZeroShotImageClassificationPipeline, SiglipProcessor, SiglipModel from torch2trt import torch2trt model = SiglipModel.from_pretrained('google/siglip-large-patch16-384', torch_dtype=torch.float16).cuda() text_model = model.text_model dummy = torch.ones(1, 64, dtype=torch.long, device='cuda') text_model(dummy)...

aliencaocao

SyntaxWarning of dataset.py:61

1

Hello. SyntaxWarning appeared when I imported torch2trt. ``` C:\DeepLearning\torch2trt-master\torch2trt\dataset.py:61: SyntaxWarning: assertion is always true, perhaps remove parentheses? assert(len(self) > 0, 'Cannot create default flattener without input data.') ``` I think...

grapevine-AI

_trt.shape overflow when converting from pytorch to TRT

5

before operator converting, some input tensors had attributes called '_trt'  To deal with it, I've deleted these incorrect '_trt' attributes manually in related converter function in “naive_converters.py”. However, I've...

Justin020718

How to save a model to TRT file? Not covered in the docs

2

I'd like to save my model into TRT file. Could you please provide a sample code?

talcs