A tutorial on pin_memory and non_blocking usage

.ci/docker/requirements.txt

@@ -28,8 +28,9 @@ tensorboard
 jinja2==3.1.3
 pytorch-lightning
 torchx
-torchrl==0.3.0
-tensordict==0.3.0
+# TODO: use stable 0.5 when released
+-e git+https://github.com/pytorch/rl.git
+-e git+https://github.com/pytorch/tensordict.git
 ax-platform
 nbformat>==5.9.2
 datasets

en-wordlist.txt

@@ -1,3 +1,4 @@
+
 ACL
 ADI
 AOT
@@ -50,6 +51,7 @@ DDP
 DDPG
 DDQN
 DLRM
+DMA
 DNN
 DQN
 DataLoaders
@@ -139,6 +141,7 @@ MKLDNN
 MLP
 MLPs
 MNIST
+MPS
 MUC
 MacBook
 MacOS
@@ -219,6 +222,7 @@ STR
 SVE
 SciPy
 Sequentials
+Sharding
 Sigmoid
 SoTA
 Sohn
@@ -254,6 +258,7 @@ VLDB
 VQA
 VS Code
 ViT
+Volterra
 WMT
 WSI
 WSIs
@@ -336,11 +341,11 @@ dataset’s
 deallocation
 decompositions
 decorrelated
-devicemesh
 deserialize
 deserialized
 desynchronization
 deterministically
+devicemesh
 dimensionality
 dir
 discontiguous
@@ -384,6 +389,7 @@ hessian
 hessians
 histoencoder
 histologically
+homonymous
 hotspot
 hvp
 hyperparameter
@@ -459,6 +465,7 @@ optimizer's
 optimizers
 otsu
 overfitting
+pageable
 parallelizable
 parallelization
 parametrization
@@ -522,7 +529,6 @@ runtime
 runtimes
 scalable
 sharded
-Sharding
 softmax
 sparsified
 sparsifier
@@ -609,4 +615,4 @@ warmstarting
 warmup
 webp
 wsi
-wsis
+wsis

index.rst

@@ -93,6 +93,13 @@ Welcome to PyTorch Tutorials
    :link: intermediate/tensorboard_tutorial.html
    :tags: Interpretability,Getting-Started,TensorBoard
 
+.. customcarditem::
+   :header: Good usage of `non_blocking` and `pin_memory()` in PyTorch
+   :card_description: A guide on best practices to copy data from CPU to GPU.
+   :image: _static/img/pinmem.png
+   :link: intermediate/pinmem_nonblock.html
+   :tags: Getting-Started
+
 .. Image/Video
 
 .. customcarditem::
@@ -942,6 +949,7 @@ Additional Resources
    beginner/basics/autogradqs_tutorial
    beginner/basics/optimization_tutorial
    beginner/basics/saveloadrun_tutorial
+   intermediate/pinmem_nonblock
    advanced/custom_ops_landing_page
 
 .. toctree::