No module named 'torch.mps'

[ahsanulhabib]

I am getting the following error while running the Kaggle notebook for detecting potholes. Is it related to the pytorch version? My current pytorch version is 2.0.0

---

ModuleNotFoundError Traceback (most recent call last) Cell In[59], line 1 ----> 1 trainer.train(model=model, 2 training_params=train_params, 3 train_loader=train_data, 4 valid_loader=val_data)

File /opt/conda/lib/python3.10/site-packages/super_gradients/training/sg_trainer/sg_trainer.py:1037, in Trainer.train(self, model, training_params, train_loader, valid_loader, additional_configs_to_log) 1034 self._prep_net_for_train() 1036 # SET RANDOM SEED -> 1037 random_seed(is_ddp=device_config.multi_gpu == MultiGPUMode.DISTRIBUTED_DATA_PARALLEL, device=device_config.device, seed=self.training_params.seed) 1039 silent_mode = self.training_params.silent_mode or self.ddp_silent_mode 1040 # METRICS

File /opt/conda/lib/python3.10/site-packages/super_gradients/training/utils/utils.py:307, in random_seed(is_ddp, device, seed) 298 """ 299 Sets random seed of numpy, torch and random. 300 (...) 304 :param seed: int, random seed to be set 305 """ 306 rank = 0 if not is_ddp else int(device.split(":")[1]) --> 307 torch.manual_seed(seed + rank) 308 np.random.seed(seed + rank) 309 random.seed(seed + rank)

File /opt/conda/lib/python3.10/site-packages/torch/random.py:42, in manual_seed(seed) 39 if not torch.cuda._is_in_bad_fork(): 40 torch.cuda.manual_seed_all(seed) ---> 42 return default_generator.manual_seed(seed)

ModuleNotFoundError: No module named 'torch.mps'

[dagshub[bot]]

Join the discussion on DagsHub!

[Louis-Dupont]

Hi @ahsanulhabib, It looks like its due to your pytorch version, you can see if this solves it: https://stackoverflow.com/questions/72535034/module-torch-has-no-attribute-has-mps

[itberrios]

I don't think pytorch 2.0 is supported yet, I had some issues previously. I haven't tried the notebook that you are referencing but so far I haven;t had much issues with these installs:

! pip install torch==1.11.0+cu113 torchvision==0.12.0+cu113 torchaudio==0.11.0 --extra-index-url https://download.pytorch.org/whl/cu113 &> /dev/null
! pip install super-gradients
! pip install pytorch-quantization==2.1.2 --extra-index-url https://pypi.ngc.nvidia.com &> /dev/null
! pip install matplotlib==3.1.3 &> /dev/null
! pip install --upgrade psutil==5.9.2 &> /dev/null
! pip install --upgrade pillow==7.1.2 &> /dev/null```

[Louis-Dupont]

Can you please write the issue you had when installing? Note that you can take out &> /dev/null to see the full install logs.

Meanwhile, you can also try to import torch and torvision again after you installed SG and see if this helps