pytorch lightning tpu on kaggle kernel error

[germayneng]

🐛 Bug

Hi - i am not able to get pytorch lightning working with TPU v3.8 on kaggle. I have followed all the instructions as well as kaggle kernel examples and it does not seem to work - at least for pytorch lightning v1.4

Error when importing lightning is as follows:

---------------------------------------------------------------------------
ModuleNotFoundError                       Traceback (most recent call last)
/tmp/ipykernel_47/4005941092.py in <module>
----> 1 import pytorch_lightning as pl

/opt/conda/lib/python3.7/site-packages/pytorch_lightning/__init__.py in <module>
     19 
     20 from pytorch_lightning import metrics  # noqa: E402
---> 21 from pytorch_lightning.callbacks import Callback  # noqa: E402
     22 from pytorch_lightning.core import LightningDataModule, LightningModule  # noqa: E402
     23 from pytorch_lightning.trainer import Trainer  # noqa: E402

/opt/conda/lib/python3.7/site-packages/pytorch_lightning/callbacks/__init__.py in <module>
     22 from pytorch_lightning.callbacks.prediction_writer import BasePredictionWriter
     23 from pytorch_lightning.callbacks.progress import ProgressBar, ProgressBarBase
---> 24 from pytorch_lightning.callbacks.pruning import ModelPruning
     25 from pytorch_lightning.callbacks.quantization import QuantizationAwareTraining
     26 from pytorch_lightning.callbacks.stochastic_weight_avg import StochasticWeightAveraging

/opt/conda/lib/python3.7/site-packages/pytorch_lightning/callbacks/pruning.py in <module>
     29 import pytorch_lightning as pl
     30 from pytorch_lightning.callbacks.base import Callback
---> 31 from pytorch_lightning.core.lightning import LightningModule
     32 from pytorch_lightning.utilities.apply_func import apply_to_collection
     33 from pytorch_lightning.utilities.distributed import rank_zero_debug, rank_zero_only

/opt/conda/lib/python3.7/site-packages/pytorch_lightning/core/__init__.py in <module>
     14 
     15 from pytorch_lightning.core.datamodule import LightningDataModule
---> 16 from pytorch_lightning.core.lightning import LightningModule
     17 
     18 __all__ = ["LightningDataModule", "LightningModule"]

/opt/conda/lib/python3.7/site-packages/pytorch_lightning/core/lightning.py in <module>
     39 from pytorch_lightning.core.optimizer import LightningOptimizer
     40 from pytorch_lightning.core.saving import ModelIO
---> 41 from pytorch_lightning.trainer.connectors.logger_connector.fx_validator import FxValidator
     42 from pytorch_lightning.utilities import rank_zero_deprecation, rank_zero_warn
     43 from pytorch_lightning.utilities.apply_func import apply_to_collection, convert_to_tensors

/opt/conda/lib/python3.7/site-packages/pytorch_lightning/trainer/__init__.py in <module>
     16 """
     17 
---> 18 from pytorch_lightning.trainer.trainer import Trainer
     19 from pytorch_lightning.utilities.seed import seed_everything
     20 

/opt/conda/lib/python3.7/site-packages/pytorch_lightning/trainer/trainer.py in <module>
     36 from pytorch_lightning.plugins import Plugin
     37 from pytorch_lightning.plugins.environments import ClusterEnvironment
---> 38 from pytorch_lightning.profiler import (
     39     AdvancedProfiler,
     40     BaseProfiler,

/opt/conda/lib/python3.7/site-packages/pytorch_lightning/profiler/__init__.py in <module>
    201 from pytorch_lightning.profiler.pytorch import PyTorchProfiler
    202 from pytorch_lightning.profiler.simple import SimpleProfiler
--> 203 from pytorch_lightning.profiler.xla import XLAProfiler
    204 
    205 __all__ = [

/opt/conda/lib/python3.7/site-packages/pytorch_lightning/profiler/xla.py in <module>
     48 
     49 if _TPU_AVAILABLE:
---> 50     import torch_xla.debug.profiler as xp
     51 
     52 log = logging.getLogger(__name__)

ModuleNotFoundError: No module named 'torch_xla.debug.profiler'

To Reproduce

I am meeting an error when i am trying to import the pytorch lightning module.

To reproduce - start a tpu kaggle kernel notebook and run:

! curl https://raw.githubusercontent.com/pytorch/xla/master/contrib/scripts/env-setup.py -o pytorch-xla-env-setup.py
! python pytorch-xla-env-setup.py --version 1.7 --apt-packages libomp5 libopenblas-dev

! pip install pytorch-lightning==1.4.0

Expected behavior

To be able to load pytorch lightning with trainer and tpu_cores enabled for training

Environment

• PyTorch Lightning Version 1.4.0
• PyTorch Version (e.g., 1.10): 1.7.0a0 / XLA: 1.7
• Python version (e.g., 3.9): 3.7.10
• OS (e.g., Linux): n/a
• CUDA/cuDNN version: cuda11
• GPU models and configuration:
• How you installed PyTorch (conda, pip, source): pip
• If compiling from source, the output of torch.__config__.show():
• Any other relevant information:

cc @kaushikb11 @tchaton

[tchaton]

Dear @germayneng,

Would you mind updating to the latest version of Lightning ?

[tchaton]

@kaushikb11

[germayneng]

still getting the same error with lightning at 1.5.3

ModuleNotFoundError: No module named 'torch_xla.debug.profiler'

I am installing i believe 1.7 of torch XLA - is this correct?

!curl https://raw.githubusercontent.com/pytorch/xla/master/contrib/scripts/env-setup.py -o pytorch-xla-env-setup.py
!python pytorch-xla-env-setup.py --version 1.7 --apt-packages libomp5 libopenblas-dev

[germayneng]

Hi any updates on this?

[githendumukiri]

any updates?