Merge branch 'main' into leslie/add_x86InductorQuantizer_Performance_Number

Author: svekars

.github/scripts/docathon-label-sync.py

@@ -25,11 +25,11 @@ def main():
     issue_number = int(re.findall(r'#(\d{1,5})', pull_request_body)[0])
     issue = repo.get_issue(issue_number)
     issue_labels = issue.labels
-    docathon_label_present = any(label.name == 'docathon-h1-2023' for label in issue_labels)
+    docathon_label_present = any(label.name == 'docathon-h2-2023' for label in issue_labels)
 
     # if the issue has a docathon label, add all labels from the issue to the PR.
     if not docathon_label_present:
-        print("The 'docathon-h1-2023' label is not present in the issue.")
+        print("The 'docathon-h2-2023' label is not present in the issue.")
         return    
     pull_request_labels = pull_request.get_labels()
     issue_label_names = [label.name for label in issue_labels]

.github/workflows/docathon-assign.yml

@@ -36,14 +36,14 @@ jobs:
                 repo: context.repo.repo,
                 issue_number: issueNumber
               });
-              const hasLabel = issue.labels.some(label => label.name === 'docathon-h1-2023');
+              const hasLabel = issue.labels.some(label => label.name === 'docathon-h2-2023');
               if (hasLabel) {
                 if (issue.assignee !== null) {
                   await octokit.issues.createComment({
                     owner: context.repo.owner,
                     repo: context.repo.repo,
                     issue_number: issueNumber,
-                    body: "The issue is already assigned. Please pick an opened and unnasigned issue with the [docathon-h1-2023 label](https://github.com/pytorch/tutorials/issues?q=is%3Aopen+is%3Aissue+label%3Adocathon-h1-2023)."
+                    body: "The issue is already assigned. Please pick an opened and unnasigned issue with the [docathon-h2-2023 label](https://github.com/pytorch/tutorials/issues?q=is%3Aopen+is%3Aissue+label%3Adocathon-h2-2023)."
                   });
                 } else {
                   octokit.issues.addAssignees({
@@ -54,7 +54,7 @@ jobs:
                   });
                 }
               } else {
-                const commmentMessage = "This issue does not have the correct label. Please pick an opened and unnasigned issue with the [docathon-h1-2023 label](https://github.com/pytorch/tutorials/issues?q=is%3Aopen+is%3Aissue+label%3Adocathon-h1-2023)."
+                const commmentMessage = "This issue does not have the correct label. Please pick an opened and unnasigned issue with the [docathon-h2-2023 label](https://github.com/pytorch/tutorials/issues?q=is%3Aopen+is%3Aissue+label%3Adocathon-h2-2023)."
                 await octokit.issues.createComment({
                   owner: context.repo.owner,
                   repo: context.repo.repo,

README.md

@@ -5,7 +5,16 @@ All the tutorials are now presented as sphinx style documentation at:
 
 ## [https://pytorch.org/tutorials](https://pytorch.org/tutorials)
 
+# Asking a question
 
+If you have a question about a tutorial, post in https://dev-discuss.pytorch.org/ rather than creating an issue in this repo. Your question will be answered much faster on the dev-discuss forum.
+
+# Submitting an issue
+
+You can submit the following types of issues:
+
+* Feature request - request a new tutorial to be added. Please explain why this tutorial is needed and how it demonstrates PyTorch value.
+* Bug report - report a failiure or outdated information in an existing tutorial. When submitting a bug report, please run: `python3 -m torch.utils.collect_env` to get information about your environment and add the output to the bug report.
 
 # Contributing
 

beginner_source/blitz/autograd_tutorial.py

@@ -321,3 +321,4 @@
 #
 # -  `In-place operations & Multithreaded Autograd <https://pytorch.org/docs/stable/notes/autograd.html>`__
 # -  `Example implementation of reverse-mode autodiff <https://colab.research.google.com/drive/1VpeE6UvEPRz9HmsHh1KS0XxXjYu533EC>`__
+# -  `Video: PyTorch Autograd Explained - In-depth Tutorial <https://www.youtube.com/watch?v=MswxJw-8PvE>`__

intermediate_source/reinforcement_ppo.py

@@ -137,7 +137,7 @@
 # actually return ``frame_skip`` frames).
 #
 
-device = "cpu" if not torch.has_cuda else "cuda:0"
+device = "cpu" if not torch.cuda.is_available() else "cuda:0"
 num_cells = 256  # number of cells in each layer i.e. output dim.
 lr = 3e-4
 max_grad_norm = 1.0

intermediate_source/torch_compile_tutorial.py

@@ -195,11 +195,15 @@ def init_model():
 # GPU compute and the observed speedup may be less significant.
 #
 # You may also see different speedup results depending on the chosen ``mode``
-# argument. Since our model and data are small, we want to reduce overhead as
-# much as possible, and so we chose ``"reduce-overhead"``. For your own models,
+# argument. The ``"reduce-overhead"`` mode uses CUDA graphs to further reduce
+# the overhead of Python. For your own models,
 # you may need to experiment with different modes to maximize speedup. You can
 # read more about modes `here <https://pytorch.org/get-started/pytorch-2.0/#user-experience>`__.
 #
+# You may might also notice that the second time we run our model with ``torch.compile`` is significantly
+# slower than the other runs, although it is much faster than the first run. This is because the ``"reduce-overhead"``
+# mode runs a few warm-up iterations for CUDA graphs.
+#
 # For general PyTorch benchmarking, you can try using ``torch.utils.benchmark`` instead of the ``timed``
 # function we defined above. We wrote our own timing function in this tutorial to show
 # ``torch.compile``'s compilation latency.

prototype_source/semi_structured_sparse.rst

@@ -315,6 +315,7 @@ Now that those are defined, we just need one additional helper function, which w
 We will get started by loading our model and tokenizer, and then setting up our dataset.
 
 .. code:: python
+
     # load model
     model_name = "bert-base-cased"
     tokenizer = transformers.AutoTokenizer.from_pretrained(model_name)
@@ -344,6 +345,7 @@ Running the following code gives me an F1 score of 86.9. This is quite close to
     training_args = transformers.TrainingArguments(
         "trainer",
         num_train_epochs=1,
+        lr_scheduler_type="constant",
         per_device_train_batch_size=64,
         per_device_eval_batch_size=512,
     )
@@ -446,7 +448,7 @@ We will also evaluate the model to show the accuracy degradation of zero-shot pr
         with torch.inference_mode():
             predictions = trainer.predict(tokenized_squad_dataset["validation"])
         pruned = compute_metrics(
-            *predictions.predictions
+            *predictions.predictions,
             tokenized_squad_dataset["validation"],
             squad_dataset["validation"],
         )
@@ -498,7 +500,7 @@ Now that we have a model in this format, we can accelerate it for inference just
     print("sparse eval metrics: ", metrics_sparse)
     sparse_perf = measure_execution_time(
         model,
-        batch_sizes_perf_cuda,
+        batch_sizes,
         tokenized_squad_dataset["validation"],
     )
     print("sparse perf metrics: ", sparse_perf)

requirements.txt

@@ -26,6 +26,7 @@ jinja2==3.0.3
 pytorch-lightning
 torchx
 torchrl==0.2.0
+tensordict==0.2.0
 ax-platform
 nbformat>=4.2.0
 datasets