Reduce number of iterations in USB tutorial (#2771)

* Add USB framework image
* Define acronyms at beginning
* Update num_train_iter to 500
* Add note on requiring CUDA
---------

Co-authored-by: atsuko <atsuko@ibm.com>
Co-authored-by: Svetlana Karslioglu <svekars@meta.com>

Author: 3 people

.jenkins/validate_tutorials_built.py

@@ -28,7 +28,6 @@
     "intermediate_source/_torch_export_nightly_tutorial",  # does not work on release
     "advanced_source/super_resolution_with_onnxruntime",
     "advanced_source/ddp_pipeline",  # requires 4 gpus
-    "advanced_source/usb_semisup_learn", # in the current form takes 140+ minutes to build - can be enabled when the build time is reduced
     "prototype_source/fx_graph_mode_ptq_dynamic",
     "prototype_source/vmap_recipe",
     "prototype_source/torchscript_freezing",

_static/img/usb_semisup_learn/code.png

@@ -5,7 +5,7 @@
 **Author**: `Hao Chen <https://github.com/Hhhhhhao>`_
 
 Unified Semi-supervised learning Benchmark (USB) is a semi-supervised
-learning framework built upon PyTorch.
+learning (SSL) framework built upon PyTorch.
 Based on Datasets and Modules provided by PyTorch, USB becomes a flexible,
 modular, and easy-to-use framework for semi-supervised learning.
 It supports a variety of semi-supervised learning algorithms, including
@@ -17,7 +17,7 @@
 This tutorial will walk you through the basics of using the USB lighting
 package.
 Let's get started by training a ``FreeMatch``/``SoftMatch`` model on
-CIFAR-10 using pretrained ViT!
+CIFAR-10 using pretrained Vision Transformers (ViT)!
 And we will show it is easy to change the semi-supervised algorithm and train
 on imbalanced datasets.
 
@@ -64,6 +64,9 @@
 # Now, let's use USB to train ``FreeMatch`` and ``SoftMatch`` on CIFAR-10.
 # First, we need to install USB package ``semilearn`` and import necessary API
 # functions from USB.
+# If you are running this in Google Colab, install ``semilearn`` by running:
+# ``!pip install semilearn``.
+#
 # Below is a list of functions we will use from ``semilearn``:
 #
 # - ``get_dataset`` to load dataset, here we use CIFAR-10
@@ -77,6 +80,10 @@
 # - ``Trainer``: a Trainer class for training and evaluating the
 # algorithm on dataset
 # 
+# Note that a CUDA-enabled backend is required for training with the ``semilearn`` package.
+# See `Enabling CUDA in Google Colab <https://pytorch.org/tutorials/beginner/colab#using-cuda>`__ for instructions
+# on enabling CUDA in Google Colab.
+#
 import semilearn
 from semilearn import get_dataset, get_data_loader, get_net_builder, get_algorithm, get_config, Trainer
 
@@ -92,7 +99,7 @@
 
     # optimization configs
     'epoch': 1,  
-    'num_train_iter': 4000,  
+    'num_train_iter': 500,
     'num_eval_iter': 500,  
     'num_log_iter': 50,  
     'optim': 'AdamW',
@@ -141,16 +148,16 @@
 
 ######################################################################
 # We can start training the algorithms on CIFAR-10 with 40 labels now.
-# We train for 4000 iterations and evaluate every 500 iterations.
+# We train for 500 iterations and evaluate every 500 iterations.
 # 
 trainer = Trainer(config, algorithm)
 trainer.fit(train_lb_loader, train_ulb_loader, eval_loader)
 
 
 ######################################################################
 # Finally, let's evaluate the trained model on the validation set.
-# After training 4000 iterations with ``FreeMatch`` on only 40 labels of
-# CIFAR-10, we obtain a classifier that achieves above 93 accuracy on the validation set.
+# After training 500 iterations with ``FreeMatch`` on only 40 labels of
+# CIFAR-10, we obtain a classifier that achieves around 87% accuracy on the validation set.
 trainer.evaluate(eval_loader)
 
 
@@ -174,7 +181,7 @@
 
     # optimization configs
     'epoch': 1,  
-    'num_train_iter': 4000,  
+    'num_train_iter': 500,
     'num_eval_iter': 500,  
     'num_log_iter': 50,  
     'optim': 'AdamW',
@@ -225,7 +232,7 @@
 
 ######################################################################
 # We can start Train the algorithms on CIFAR-10 with 40 labels now.
-# We train for 4000 iterations and evaluate every 500 iterations.
+# We train for 500 iterations and evaluate every 500 iterations.
 # 
 trainer = Trainer(config, algorithm)
 trainer.fit(train_lb_loader, train_ulb_loader, eval_loader)
@@ -239,8 +246,8 @@
 
 
 ######################################################################
-# References
-# [1] USB: https://github.com/microsoft/Semi-supervised-learning
-# [2] Kihyuk Sohn et al. FixMatch: Simplifying Semi-Supervised Learning with Consistency and Confidence
-# [3] Yidong Wang et al. FreeMatch: Self-adaptive Thresholding for Semi-supervised Learning
-# [4] Hao Chen et al. SoftMatch: Addressing the Quantity-Quality Trade-off in Semi-supervised Learning
+# References:
+# - [1] USB: https://github.com/microsoft/Semi-supervised-learning
+# - [2] Kihyuk Sohn et al. FixMatch: Simplifying Semi-Supervised Learning with Consistency and Confidence
+# - [3] Yidong Wang et al. FreeMatch: Self-adaptive Thresholding for Semi-supervised Learning
+# - [4] Hao Chen et al. SoftMatch: Addressing the Quantity-Quality Trade-off in Semi-supervised Learning

advanced_source/usb_semisup_learn.py

@@ -93,3 +93,11 @@ Hopefully this example will give you a good starting point for running
 some of the more complex tutorials in Colab. As we evolve our use of
 Colab on the PyTorch tutorials site, we'll look at ways to make this
 easier for users.
+
+Enabling CUDA
+~~~~~~~~~~~~~~~~
+Some tutorials require a CUDA-enabled device (NVIDIA GPU), which involves
+changing the Runtime type prior to executing the tutorial.
+To change the Runtime in Google Colab, on the top drop-down menu select **Runtime**,
+then select **Change runtime type**. Under **Hardware accelerator**, select ``T4 GPU``,
+then click ``Save``.