Merge pull request #114 from mila-iqia/updating_torch

Updating torch and others

.gitignore

@@ -2,9 +2,10 @@
 mlruns
 
 examples/data/
-examples/*/output/
+examples/*/output*/
+examples/*/orion_working_dir*/
 examples/*/lightning_logs/
-
+examples/*/orion_db.pkl
 # Byte-compiled / optimized / DLL files
 __pycache__/
 *.py[cod]
@@ -115,4 +116,4 @@ venv.bak/
 *.swo
 
 # vscode
-.vscode
+.vscode

amlrt_project/train.py

@@ -212,6 +212,8 @@ def train_impl(model, datamodule, output, hyper_params, use_progress_bar, gpus):
  # Log the best result and associated hyper parameters
  best_dev_result = float(early_stopping.best_score.cpu().numpy())
  logger.log_hyperparams(hyper_params, metrics={'best_dev_metric': best_dev_result})
+ with open(os.path.join(output, 'results.txt'), 'w') as stream_out:
+ stream_out.write(f'final best_dev_metric: {best_dev_result}\n')
 
  return best_dev_result
 

examples/local/config.yaml

@@ -2,7 +2,7 @@
 batch_size: 32
 optimizer: adam
 loss: cross_entropy
-max_epoch: 5
+max_epoch: 1
 exp_name: my_exp_1
 num_workers: 0
 # set to null to avoid setting a seed (can speed up GPU computation, but

examples/local_orion/config.yaml

@@ -2,7 +2,7 @@
 batch_size: 32
 optimizer: adam
 loss: cross_entropy
-max_epoch: 5
+max_epoch: 1
 exp_name: my_exp_1
 num_workers: 0
 # set to null to avoid setting a seed (can speed up GPU computation, but

examples/local_orion/run.sh

@@ -1,4 +1,3 @@
-export MLFLOW_TRACKING_URI='mlruns'
 export ORION_DB_ADDRESS='orion_db.pkl'
 export ORION_DB_TYPE='pickleddb'
 

setup.py

@@ -17,15 +17,16 @@
  'pytest==7.1.2',
  'pytest-cov==3.0.0',
  'pytorch_lightning==1.8.3',
- 'pytype==2023.1.17',
+ 'pytype==2024.2.27',
  'sphinx==7.2.6',
  'sphinx-autoapi==3.0.0',
  'sphinx-rtd-theme==1.3.0',
  'sphinxcontrib-napoleon==0.7',
  'sphinxcontrib-katex==0.9.9',
+ 'tensorboard==2.16.2',
  'tqdm==4.64.0',
- 'torch==1.12.0',
- 'torchvision==0.13.0',
+ 'torch==2.2.1',
+ 'torchvision==0.17.1',
  ],
  entry_points={
  'console_scripts': [

tests/end2end_pytorch/run.sh

@@ -4,11 +4,13 @@ set -e
 # go to the examples folder and run the example
 cd $GITHUB_WORKSPACE/examples/local
 sh run.sh
-mv output outout_OLD
+mv output output_OLD
 # re-run the example to check reproducibility
 sh run.sh
 # check results are the same
-DIFF_LINES=`grep "best_dev_metric" output*/stats.yaml | sed 's@^.*best_dev_metric: @@g' | uniq | wc -l`
+echo "results are:"
+cat output*/results.txt
+DIFF_LINES=`cat output*/results.txt | uniq | wc -l`
 if [ ${DIFF_LINES} -gt 1 ]; then
  echo "ERROR: two identical runs produced different output results - review seed implementation"
  exit 1
@@ -23,8 +25,11 @@ mv orion_working_dir orion_working_dir_OLD
 # re-run the example to check reproducibility
 rm -fr orion_db*
 sh run.sh
-DIFF_LINES=`grep "best_dev_metric" orion_working_dir*/*/stats.yaml | sed 's@^.*best_dev_metric: @@g' | sort | uniq | wc -l`
-if [ ${DIFF_LINES} -gt 2 ]; then
+# check results are the same
+echo "results are:"
+cat orion_working_dir*/*/results.txt
+DIFF_LINES=`grep "best_dev_metric" orion_working_dir*/*/results.txt | sed 's@^.*best_dev_metric: @@g' | sort | uniq | wc -l`
+if [ ${DIFF_LINES} -gt 2 ]; then # note we have two trials per experiment, this is why we can have 2 different results - but not more
  echo "ERROR: two identical Orion runs produced different output results - review seed implementation"
  exit 1
 else