CI: update transformer engine & 1-bit adam DeepSpeed links in documentation and code comments

.github/CONTRIBUTING.md

@@ -103,7 +103,7 @@ _**Note**, even if you do not find the solution, sending a PR with a test coveri
 
 Want to keep Lightning healthy? Love seeing those green tests? So do we! How to we keep it that way? We write tests! We value tests contribution even more than new features.
 
-Most of the tests in PyTorch Lightning train a random `BoringModel` under various trainer conditions (ddp, amp, etc...). Want to add a new test case and not sure how? [Talk to us!](https://www.pytorchlightning.ai/community)
+Most of the tests in PyTorch Lightning train a random `BoringModel` under various trainer conditions (ddp, amp, etc...). Want to add a new test case and not sure how? [Talk to us!](https://www.lightning.ai/community)
 
 ______________________________________________________________________
 

docs/source-fabric/_templates/theme_variables.jinja

@@ -9,7 +9,7 @@
   'get_started': '',
   'blog': 'https://lightning.ai/pages/blog/',
   'support': '',
-  'community': 'https://www.pytorchlightning.ai/community',
+  'community': 'https://www.lightning.ai/community',
   'forums': 'https://lightning.ai/forums/',
 }
 -%}

docs/source-fabric/fundamentals/precision.rst

@@ -139,7 +139,7 @@ memory utilization in both training and inference. It offers improved performanc
 
 Using TE requires replacing some of the layers in your model. Fabric automatically replaces the :class:`torch.nn.Linear`
 and :class:`torch.nn.LayerNorm` layers in your model with their TE alternatives, however, TE also offers
-`fused layers <https://docs.nvidia.com/deeplearning/transformer-engine/user-guide/api/pytorch.html>`__
+`fused layers <https://docs.nvidia.com/deeplearning/transformer-engine/api/pytorch.html>`__
 to squeeze out all the possible performance. If Fabric detects that any layer has been replaced already, automatic
 replacement is not done.
 
@@ -162,7 +162,7 @@ the model and inputs can be kept in true full or half precision.
     fabric = Fabric(plugins=precision)
 
 
-Under the hood, we use `transformer_engine.pytorch.fp8_autocast <https://docs.nvidia.com/deeplearning/transformer-engine/user-guide/api/pytorch.html#transformer_engine.pytorch.fp8_autocast>`__ with the default fp8 recipe.
+Under the hood, we use `transformer_engine.pytorch.fp8_autocast <https://docs.nvidia.com/deeplearning/transformer-engine/api/pytorch.html#transformer_engine.pytorch.fp8_autocast>`__ with the default fp8 recipe.
 
 .. note::
 

docs/source-pytorch/_templates/theme_variables.jinja

@@ -5,15 +5,15 @@
   'governance': 'https://lightning.ai/docs/pytorch/latest/community/governance.html',
   'docs': 'https://lightning.ai/docs/pytorch/latest/',
   'twitter': 'https://twitter.com/LightningAI',
-  'discuss': 'https://www.pytorchlightning.ai/community',
+  'discuss': 'https://www.lightning.ai/community',
   'tutorials': 'https://lightning.ai/docs/pytorch/latest/#tutorials',
   'home': 'https://lightning.ai/docs/pytorch/latest/',
   'get_started': 'https://lightning.ai/docs/pytorch/latest/starter/introduction.html',
   'features': 'https://lightning.ai/docs/pytorch/latest/',
   'blog': 'https://lightning.ai/pages/blog/',
   'resources': 'https://lightning.ai/docs/pytorch/latest/#community-examples',
   'support': 'https://lightning.ai/docs/pytorch/latest/',
-  'community': 'https://www.pytorchlightning.ai/community',
+  'community': 'https://www.lightning.ai/community',
   'forums': 'https://lightning.ai/forums/',
   'versions': 'https://lightning.ai/docs/pytorch/latest/past_versions.html'
 }

docs/source-pytorch/advanced/model_parallel/deepspeed.rst

@@ -8,7 +8,7 @@ DeepSpeed
 
 `DeepSpeed <https://github.com/microsoft/DeepSpeed>`__ is a deep learning training optimization library, providing the means to train massive billion parameter models at scale.
 Using the DeepSpeed strategy, we were able to **train model sizes of 10 Billion parameters and above**, with a lot of useful information in this `benchmark <https://github.com/huggingface/transformers/issues/9996>`_ and the `DeepSpeed docs <https://www.deepspeed.ai/tutorials/megatron/>`__.
-DeepSpeed also offers lower level training optimizations, and efficient optimizers such as `1-bit Adam <https://www.deepspeed.ai/tutorials/onebit-adam/>`_. We recommend using DeepSpeed in environments where speed and memory optimizations are important (such as training large billion parameter models).
+DeepSpeed also offers lower level training optimizations, and efficient optimizers. We recommend using DeepSpeed in environments where speed and memory optimizations are important (such as training large billion parameter models).
 
 .. warning::  This is an :ref:`experimental <versioning:Experimental API>` feature.
 

docs/source-pytorch/common/precision_intermediate.rst

@@ -128,7 +128,7 @@ memory utilization in both training and inference. It offers improved performanc
 
 Using TE requires replacing some of the layers in your model. Fabric automatically replaces the :class:`torch.nn.Linear`
 and :class:`torch.nn.LayerNorm` layers in your model with their TE alternatives, however, TE also offers
-`fused layers <https://docs.nvidia.com/deeplearning/transformer-engine/user-guide/api/pytorch.html>`__
+`fused layers <https://docs.nvidia.com/deeplearning/transformer-engine/api/pytorch.html>`__
 to squeeze out all the possible performance. If Fabric detects that any layer has been replaced already, automatic
 replacement is not done.
 
@@ -151,7 +151,7 @@ the model and inputs can be kept in true full or half precision.
     trainer = Trainer(plugins=precision)
 
 
-Under the hood, we use `transformer_engine.pytorch.fp8_autocast <https://docs.nvidia.com/deeplearning/transformer-engine/user-guide/api/pytorch.html#transformer_engine.pytorch.fp8_autocast>`__ with the default fp8 recipe.
+Under the hood, we use `transformer_engine.pytorch.fp8_autocast <https://docs.nvidia.com/deeplearning/transformer-engine/api/pytorch.html#transformer_engine.pytorch.fp8_autocast>`__ with the default fp8 recipe.
 
 .. note::
 

examples/pytorch/bug_report/bug_report_model.ipynb

@@ -45,7 +45,7 @@
         "# The Boring Model\n",
         "Replicate a bug you experience, using this model.\n",
         "\n",
-        "[Remember! we're always available for support on Slack](https://www.pytorchlightning.ai/community)"
+        "[Remember! we're always available for support on Slack](https://www.lightning.ai/community)"
       ]
     },
     {

src/lightning/fabric/plugins/precision/transformer_engine.py

@@ -46,7 +46,7 @@ class TransformerEnginePrecision(Precision):
     Args:
         weights_dtype: The weights dtype to use.
         recipe: Recipe for the DelayedScaling
-            `configuration <https://docs.nvidia.com/deeplearning/transformer-engine/user-guide/api/common.html#transformer_engine.common.recipe.DelayedScaling>`__.
+            `configuration <https://docs.nvidia.com/deeplearning/transformer-engine/api/common.html#transformer_engine.common.recipe.DelayedScaling>`__.
             In dict format or the dataclass format.
         replace_layers: Whether to replace ``Linear`` and ``LayerNorm`` layers automatically with their Transformer
             Engine alternatives. Note that they don't subclass the torch equivalents so checks like
@@ -154,7 +154,7 @@ def _convert_layers(module: torch.nn.Module) -> None:
     for name, child in module.named_children():
         if isinstance(child, torch.nn.Linear):
             if child.in_features % 8 != 0 or child.out_features % 16 != 0:
-                # https://docs.nvidia.com/deeplearning/transformer-engine/user-guide/examples/fp8_primer.html#FP8-autocasting
+                # https://docs.nvidia.com/deeplearning/transformer-engine/examples/fp8_primer.html#FP8-autocasting
                 rank_zero_warn(
                     "Support for FP8 in the linear layers with this plugin is currently limited to"
                     " tensors with shapes where the dimensions are divisible by 8 and 16 respectively."
@@ -178,5 +178,5 @@ def _convert_layers(module: torch.nn.Module) -> None:
             module.__setattr__(name, replacement)
         else:
             # there are other transformer engine layers that we could convert but require fusion. full list at:
-            # https://docs.nvidia.com/deeplearning/transformer-engine/user-guide/api/pytorch.html
+            # https://docs.nvidia.com/deeplearning/transformer-engine/api/pytorch.html
             _convert_layers(child)

src/lightning/pytorch/plugins/precision/transformer_engine.py

@@ -24,7 +24,7 @@ class TransformerEnginePrecision(Precision, FabricTEPrecision):
     Args:
         dtype: The weights dtype to use.
         recipe: Recipe for the DelayedScaling
-            `configuration <https://docs.nvidia.com/deeplearning/transformer-engine/user-guide/api/common.html#transformer_engine.common.recipe.DelayedScaling>`__.
+            `configuration <https://docs.nvidia.com/deeplearning/transformer-engine/api/common.html#transformer_engine.common.recipe.DelayedScaling>`__.
             In dict format or the dataclass format.
         replace_layers: Whether to replace ``Linear`` and ``LayerNorm`` layers automatically with their Transformer
             Engine alternatives. Note that they don't subclass the torch equivalents so checks like

tests/README.md

@@ -1,6 +1,6 @@
 # PyTorch-Lightning Tests
 
-Most of the tests in PyTorch Lightning train a [BoringModel](https://github.com/Lightning-AI/pytorch-lightning/blob/master/src/lightning/pytorch/demos/boring_classes.py) under various trainer conditions (ddp, amp, etc...). Want to add a new test case and not sure how? [Talk to us!](https://www.pytorchlightning.ai/community)
+Most of the tests in PyTorch Lightning train a [BoringModel](https://github.com/Lightning-AI/pytorch-lightning/blob/master/src/lightning/pytorch/demos/boring_classes.py) under various trainer conditions (ddp, amp, etc...). Want to add a new test case and not sure how? [Talk to us!](https://www.lightning.ai/community)
 
 ## Running tests