Updated the examples to use pytorch 1.9.0 and pytorch-lightning 1.3.5 (…

…mlflow#4458)

* Updated the examples to use pytorch 1.9.0 and pytorch-lightning 1.3.5

Signed-off-by: KasirajanA <kasirajan.alayappan@ideas2it.com>
Signed-off-by: Kasirajan Alayappan <kasirajan.alayappan@ideas2it.com>

* Pytorch bert example fix for gpu run

Signed-off-by: Ubuntu <ubuntu@ip-172-31-26-167.us-east-2.compute.internal>
Signed-off-by: Kasirajan Alayappan <kasirajan.alayappan@ideas2it.com>

* Updated the torchvision version in pytorch examples

Signed-off-by: KasirajanA <kasirajan.alayappan@ideas2it.com>
Signed-off-by: Kasirajan Alayappan <kasirajan.alayappan@ideas2it.com>

* Skipping pytorch override pickle module testcase for torch>=1.9.0

Signed-off-by: Kasirajan Alayappan <kasirajan.alayappan@ideas2it.com>

* Skipping pytorch override pickle module testcase for torch>=1.9.0

Signed-off-by: Kasirajan Alayappan <kasirajan.alayappan@ideas2it.com>

* Revert "Skipping pytorch override pickle module testcase for torch>=1.9.0"

This reverts commit 7da47f8.

Signed-off-by: Kasirajan Alayappan <kasirajan.alayappan@ideas2it.com>

* Revert "Skipping pytorch override pickle module testcase for torch>=1.9.0"

This reverts commit 6277430.

Signed-off-by: Kasirajan Alayappan <kasirajan.alayappan@ideas2it.com>

* Addressing review comments

Signed-off-by: Shrinath Suresh <shrinath@ideas2it.com>

* Removing unused columns

Signed-off-by: Shrinath Suresh <shrinath@ideas2it.com>

* Setting default sample size to 2000

Signed-off-by: Shrinath Suresh <shrinath@ideas2it.com>

Co-authored-by: Ubuntu <ubuntu@ip-172-31-26-167.us-east-2.compute.internal>
Co-authored-by: Shrinath Suresh <shrinath@ideas2it.com>

examples/pytorch/AxHyperOptimizationPTL/conda.yaml

@@ -5,7 +5,7 @@ dependencies:
 - pip
 - pip:
   - mlflow
-  - pytorch-lightning
+  - pytorch-lightning==1.3.5
   - ax-platform
-  - torch>=1.6.0
-  - torchvision
+  - torchvision>=0.9.1
+  - torch==1.9.0

examples/pytorch/BertNewsClassification/MLproject

@@ -11,7 +11,7 @@ entry_points:
       batch_size: {type: int, default: 64}
       num_workers: {type: int, default: 3}
       learning_rate: {type: float, default: 0.001}
-      num_samples: {type: int, default: 15000}
+      num_samples: {type: int, default: 2000}
       dataset: {type: str, default: "20newsgroups"}
 
     command: |

examples/pytorch/BertNewsClassification/bert_classification.py

@@ -21,8 +21,7 @@
 from torch import nn
 from torch.utils.data import Dataset, DataLoader
 from transformers import BertModel, BertTokenizer, AdamW
-from torchtext.utils import download_from_url, extract_archive
-from torchtext.datasets.text_classification import URLS
+import torchtext.datasets as td
 
 
 def get_20newsgroups(num_samples):
@@ -32,9 +31,9 @@ def get_20newsgroups(num_samples):
 
 
 def get_ag_news(num_samples):
-    dataset_tar = download_from_url(URLS["AG_NEWS"], root=".data")
-    extracted_files = extract_archive(dataset_tar)
-    train_csv_path = list(filter(lambda x: x.endswith("train.csv"), extracted_files))[0]
+    # reading the input
+    td.AG_NEWS(root="data", split=("train", "test"))
+    train_csv_path = "data/AG_NEWS/train.csv"
     return (
         pd.read_csv(train_csv_path, usecols=[0, 2], names=["label", "description"])
         .assign(label=lambda df: df["label"] - 1)  # make labels zero-based
@@ -401,12 +400,12 @@ def configure_optimizers(self):
     early_stopping = EarlyStopping(monitor="val_loss", mode="min", verbose=True)
 
     checkpoint_callback = ModelCheckpoint(
-        filepath=os.getcwd(), save_top_k=1, verbose=True, monitor="val_loss", mode="min", prefix="",
+        dirpath=os.getcwd(), save_top_k=1, verbose=True, monitor="val_loss", mode="min",
     )
     lr_logger = LearningRateMonitor()
 
     trainer = pl.Trainer.from_argparse_args(
-        args, callbacks=[lr_logger, early_stopping], checkpoint_callback=checkpoint_callback
+        args, callbacks=[lr_logger, early_stopping, checkpoint_callback], checkpoint_callback=True
     )
     trainer.fit(model, dm)
     trainer.test()

examples/pytorch/BertNewsClassification/conda.yaml

@@ -10,7 +10,7 @@ dependencies:
   - boto3
   - transformers>=4.0.0
   - pandas
-  - torch==1.6.0
-  - torchvision==0.7.0
-  - torchtext==0.7.0
-  - pytorch-lightning==1.0.2
+  - torchvision>=0.9.1
+  - torch==1.9.0
+  - torchtext==0.10.0
+  - pytorch-lightning==1.3.5

examples/pytorch/IterativePruning/conda.yaml

@@ -5,9 +5,9 @@ dependencies:
 - pip
 - pip:
   - mlflow
-  - torchvision
+  - torchvision>=0.9.1
   - cloudpickle==1.6.0
-  - pytorch_lightning>=1.0.5
+  - pytorch_lightning==1.3.5
   - ax-platform
   - prettytable
-  - torch>=1.6.0
+  - torch==1.9.0

examples/pytorch/MNIST/conda.yaml

@@ -5,6 +5,6 @@ dependencies:
 - pip
 - pip:
   - mlflow
-  - torch==1.8.1
-  - torchvision==0.9.1
-  - pytorch-lightning==1.0.2
+  - torchvision>=0.9.1
+  - torch==1.9.0
+  - pytorch-lightning==1.3.5

examples/pytorch/MNIST/mnist_autolog_example.py

@@ -283,12 +283,12 @@ def configure_optimizers(self):
     )
 
     checkpoint_callback = ModelCheckpoint(
-        filepath=os.getcwd(), save_top_k=1, verbose=True, monitor="val_loss", mode="min", prefix="",
+        dirpath=os.getcwd(), save_top_k=1, verbose=True, monitor="val_loss", mode="min"
     )
     lr_logger = LearningRateMonitor()
 
     trainer = pl.Trainer.from_argparse_args(
-        args, callbacks=[lr_logger, early_stopping], checkpoint_callback=checkpoint_callback
+        args, callbacks=[lr_logger, early_stopping, checkpoint_callback], checkpoint_callback=True
     )
     trainer.fit(model, dm)
     trainer.test()

examples/pytorch/torchscript/IrisClassification/conda.yaml

@@ -7,5 +7,5 @@ dependencies:
   - sklearn
   - cloudpickle==1.6.0
   - boto3
-  - torch==1.6.0
-  - torchvision==0.7.0
+  - torchvision>=0.9.1
+  - torch==1.9.0

examples/pytorch/torchscript/MNIST/conda.yaml

@@ -7,5 +7,5 @@ dependencies:
   - mlflow
   - cloudpickle==1.6.0
   - boto3
-  - torch==1.6.0
-  - torchvision==0.7.0
+  - torchvision>=0.9.1
+  - torch==1.9.0