5090 Support

[rikabi89]

My understanding is this repo require pytorch 2.6.0 whereas the 5090 only support 2.7.0 which is not compatible with this repos. Any work around?
RuntimeError: fairseq2 requires a CUDA 12.4 build of PyTorch 2.6.0, but the installed version is a CUDA 12.8 build of PyTorch 2.7.0. Either follow the instructions at https://pytorch.org/get-started/locally to update PyTorch, or the instructions at https://github.com/facebookresearch/fairseq2#variants to update fairseq2.

[bryananderson]

As you can see in that fairseq2 link, Torch 2.7 support is available when building the package from main. You could try editing requirements.txt to build fairseq2 from main, as well as updating the relevant Torch-related packages.

[YavorGIvanov]

Yup. Checkout the fairseq2 README.

Here is a relevant part:

`

Variants

Besides PyPI, fairseq2 also has pre-built packages available for different PyTorch versions hosted on FAIR's package repository. The following matrix shows the supported combinations.

fairseq2	PyTorch	Python	Arch
HEAD	2.7.0	>=3.10, <=3.12	arm64
0.4	2.6.0	>=3.10, <=3.12	arm64

To install a specific combination, first follow the installation instructions on pytorch.org for the desired PyTorch version, and then use the following command (shown for PyTorch 2.7.0):

pip install fairseq2

--extra-index-url https://fair.pkg.atmeta.com/fairseq2/whl/pt2.7.0/cpu

Variants
Besides PyPI, fairseq2 also has pre-built packages available for different PyTorch versions hosted on FAIR's package repository. The following matrix shows the supported combinations.

fairseq2 PyTorch Python Arch
HEAD 2.7.0 >=3.10, <=3.12 arm64
0.4 2.6.0 >=3.10, <=3.12 arm64
To install a specific combination, first follow the installation instructions on pytorch.org for the desired PyTorch version, and then use the following command (shown for PyTorch 2.7.0):

pip install fairseq2
--extra-index-url https://fair.pkg.atmeta.com/fairseq2/whl/pt2.7.0/cpu`

[rikabi89]

I been trying the above for a while, not sure where I'm going wrong but I keep getting mismatch issues :```
Traceback (most recent call last):
File "/mnt/d/PLaydiffusion/app.py", line 11, in
inpainter = PlayDiffusion()
^^^^^^^^^^^^^^^
File "/root/anaconda3/envs/playdiffusion/lib/python3.11/site-packages/playdiffusion/inference.py", line 47, in init
self.mm = PlayDiffusionModelManager(self.preset, self.device)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/root/anaconda3/envs/playdiffusion/lib/python3.11/site-packages/playdiffusion/models/model_manager.py", line 17, in init
self.load_tokenizers(
File "/root/anaconda3/envs/playdiffusion/lib/python3.11/site-packages/playdiffusion/models/model_manager.py", line 44, in load_tokenizers
from playdiffusion.models.speech_tokenizer.speech_tokenizer import SpeechTokenizer
File "/root/anaconda3/envs/playdiffusion/lib/python3.11/site-packages/playdiffusion/models/speech_tokenizer/speech_tokenizer.py", line 5, in
from fairseq2.data import Collater
File "/mnt/d/PLaydiffusion/fairseq2/src/fairseq2/init.py", line 11, in
import fairseq2n # Report any fairseq2n initialization error eagerly.
^^^^^^^^^^^^^^^^
File "/root/anaconda3/envs/playdiffusion/lib/python3.11/site-packages/fairseq2n/init.py", line 194, in
_check_torch_version()
File "/root/anaconda3/envs/playdiffusion/lib/python3.11/site-packages/fairseq2n/init.py", line 189, in _check_torch_version
raise RuntimeError(
RuntimeError: fairseq2 requires a CUDA 12.4 build of PyTorch 2.6.0, but the installed version is a CUDA 12.8 build of PyTorch 2.7.0. Either follow the instructions at https://pytorch.org/get-

[bryananderson]

Looks like you are still installing the torch2.6 version of fairseq2. On their readme, under the version matrix, it shows how to install the exact correct version for your torch/cuda version.

[rikabi89]

Looks like you are still installing the torch2.6 version of fairseq2. On their readme, under the version matrix, it shows how to install the exact correct version for your torch/cuda version.

even with 2.7.0 it didn't work, I chat GPT is suggesting build pytorch from the source as well, but I think this beyond my me, so I will wait for some kinda of fork.

> ; Successfully installed nvidia-cublas-cu12-12.6.4.1 nvidia-cuda-cupti-cu12-12.6.80 nvidia-cuda-nvrtc-cu12-12.6.77 nvidia-cuda-runtime-cu12-12.6.77 nvidia-cudnn-cu12-9.5.1.17 nvidia-cufft-cu12-11.3.0.4 nvidia-curand-cu12-10.3.7.77 nvidia-cusolver-cu12-11.7.1.2 nvidia-cusparse-cu12-12.5.4.2 nvidia-cusparselt-cu12-0.6.3 nvidia-nccl-cu12-2.26.2 nvidia-nvjitlink-cu12-12.6.85 nvidia-nvtx-cu12-12.6.77 sympy-1.13.3 torch-2.7.0+cu126 torchaudio-2.7.0+cu126 triton-3.3.0
> WARNING: Running pip as the 'root' user can result in broken permissions and conflicting behaviour with the system package manager, possibly rendering your system unusable. It is recommended to use a virtual environment instead: https://pip.pypa.io/warnings/venv. Use the --root-user-action option if you know what you are doing and want to suppress this warning.
> (playdiffusion) root@DESKTOP-G461NBO:/mnt/d/PlayDiffusion/fairseq2# python -c "import fairseq2; print(fairseq2.__version__)"
> 0.5.0.dev0
> (playdiffusion) root@DESKTOP-G461NBO:/mnt/d/PlayDiffusion/fairseq2# conda activate playdiffusion
> d:
> cd /mnt/d/PLaydiffusion
>  python app.py
> 
> d:: command not found
> 
> 
> 
> ^Z
> [1]+  Stopped                 python app.py
> (playdiffusion) root@DESKTOP-G461NBO:/mnt/d/PLaydiffusion#
> (playdiffusion) root@DESKTOP-G461NBO:/mnt/d/PLaydiffusion#
> (playdiffusion) root@DESKTOP-G461NBO:/mnt/d/PLaydiffusion#  python app.py
> /root/anaconda3/envs/playdiffusion/lib/python3.11/site-packages/torch/cuda/__init__.py:287: UserWarning:
> NVIDIA GeForce RTX 5090 with CUDA capability sm_120 is not compatible with the current PyTorch installation.
> The current PyTorch install supports CUDA capabilities sm_50 sm_60 sm_70 sm_75 sm_80 sm_86 sm_90.
> If you want to use the NVIDIA GeForce RTX 5090 GPU with PyTorch, please check the instructions at https://pytorch.org/get-started/locally/
> 

[bryananderson]

I don't think this is right; the line you pasted before shows that you are using the fairseq2 version which is meant for torch2.6:

RuntimeError: fairseq2 requires a CUDA 12.4 build of PyTorch 2.6.0, but the installed version is a CUDA 12.8 build of PyTorch 2.7.0

And the fairseq2 readme shows how to install the torch2.7 version:

To install a specific combination, first follow the installation instructions on pytorch.org for the desired PyTorch version, and then use the following command (shown for PyTorch 2.7.0 and variant cu124):

pip install fairseq2\
  --extra-index-url https://fair.pkg.atmeta.com/fairseq2/whl/pt2.7.0/cu126

[dconsorte]

If you're still stuck, here's a repo to get PyTorch and Tensorflow up and running for LLM model training in a Docker container, tested on Ubuntu 24.04.2 LTS with an NVidia RTX 5090 and 5060 GPU: https://github.com/dconsorte/pytorch-tensorflow-gpu