ComfyUi error float8_e4m3fn with Pytorch

Robotics & Edge Computing Jetson & Embedded Systems Jetson AGX Xavier
1

hello i have Jetson AGX Xavier
installed torch-2.1.0a0+41361538.nv23.06-cp38-cp38-linux_aarch64.whl
and torchvision
git clone --branch v0.16.1 GitHub - pytorch/vision: Datasets, Transforms and Models specific to Computer Vision torchvision
python3 setup.py install --user

but when i run ComfyUi workflow (Flux)
i have always this error AttributeError: module ‘torch’ has no attribute ‘float8_e4m3fn’

tried several things, but cant find how to fix it (tried even install CUDA 11.8), but still same error
any idea how to fix its ?

3

Hi,

Based on the below link, it seems to be a compatibility issue with the transformers package.
Could you try the suggestion to see if it helps?

Thanks.

4

thanks tried
transformers-4.43.2
transformers-4.42.4
but still module ‘torch’ has no attribute ‘float8_e4m3fn’
any idea how to fix it ?

can we use newer version pytorch…?
looks like 2.1 is not well supported

5

Hi,

The latest PyTorch prebuilt we provided for Jetson is v2.1.
But you can build it from the source to get a newer version.

The below topic contains the building instructions for your reference:

Thanks.

6

thanks tried to compile pytorch 2.4.0 manually but still without cuda support… any idea whats wrong ?
here is end of script looks like cuda is detected but in python its not working

running build_ext
-- Building with NumPy bindings
-- Detected cuDNN at ,
-- Detected CUDA at /usr/local/cuda
-- Not using MKLDNN
-- Building NCCL library
-- Building with distributed package:
  -- USE_TENSORPIPE=True
  -- USE_GLOO=True
  -- USE_MPI=False
-- Building Executorch
-- Not using ITT
Copying functorch._C from functorch/functorch.so to /home/jan/pytorch/build/lib.linux-aarch64-cpython-38/functorch/_C.cpython-38-aarch64-linux-gnu.so
running install_lib
copying build/lib.linux-aarch64-cpython-38/torch/_C.cpython-38-aarch64-linux-gnu.so -> /usr/lib/python3.8/site-packages/torch
copying build/lib.linux-aarch64-cpython-38/torch/_export/db/examples/user_input_mutation.py -> /usr/lib/python3.8/site-packages/torch/_export/db/examples
copying build/lib.linux-aarch64-cpython-38/functorch/dim/magic_trace.py -> /usr/lib/python3.8/site-packages/functorch/dim
copying build/lib.linux-aarch64-cpython-38/torchgen/packaged/ATen/native/native_functions.yaml -> /usr/lib/python3.8/site-packages/torchgen/packaged/ATen/native
copying build/lib.linux-aarch64-cpython-38/torchgen/packaged/ATen/native/tags.yaml -> /usr/lib/python3.8/site-packages/torchgen/packaged/ATen/native
running install_egg_info
running egg_info
writing torch.egg-info/PKG-INFO
writing dependency_links to torch.egg-info/dependency_links.txt
writing entry points to torch.egg-info/entry_points.txt
writing requirements to torch.egg-info/requires.txt
writing top-level names to torch.egg-info/top_level.txt
reading manifest file 'torch.egg-info/SOURCES.txt'
reading manifest template 'MANIFEST.in'
warning: no previously-included files matching '*.o' found anywhere in distribution
warning: no previously-included files matching '*.dylib' found anywhere in distribution
warning: no previously-included files matching '*.swp' found anywhere in distribution
adding license file 'LICENSE'
adding license file 'NOTICE'
writing manifest file 'torch.egg-info/SOURCES.txt'
removing '/usr/lib/python3.8/site-packages/torch-2.2.0a0+git8ac9b20-py3.8.egg-info' (and everything under it)
Copying torch.egg-info to /usr/lib/python3.8/site-packages/torch-2.2.0a0+git8ac9b20-py3.8.egg-info
running install_scripts
Installing convert-caffe2-to-onnx script to /usr/bin
Installing convert-onnx-to-caffe2 script to /usr/bin
Installing torchrun script to /usr/bin
7

ok found solution
with this pytorch 2.2 jp5/cu114 index error is gone…
(its compiled with cuda 11.4)

10

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.