[Docs] Fix broken links (#3597)

Signed-off-by: Santosh Bhavani <santosh.bhavani@live.com>

Author: sbhavani

README.rst

@@ -38,7 +38,7 @@ News
 ====
 
 * **[09/2026]** `Transformer Engine v2.19 <https://github.com/NVIDIA/TransformerEngine/releases/tag/v2.19>`_ adds Rubin support, hybrid quantization, MXFP8 EP communication, and expanded FP8 attention support.
-* **[09/2026]** `Accelerating Dropless MoE Training in JAX with NVIDIA Transformer Engine <https://developer.nvidia.com/blog/accelerating-dropless-moe-training-in-jax-with-nvidia-transform-engine/>`_ describes optimized JAX MoE training on GB200 and GB300 systems.
+* **[09/2026]** `Accelerating Dropless MoE Training in JAX with NVIDIA Transformer Engine <https://developer.nvidia.com/blog/accelerating-dropless-moe-training-in-jax-with-nvidia-transformer-engine/>`_ describes optimized JAX MoE training on GB200 and GB300 systems.
 * **[08/2026]** `Transformer Engine v2.18 <https://github.com/NVIDIA/TransformerEngine/releases/tag/v2.18>`_ adds FP8 block scaling in PyTorch, zero-copy EP, and CUDA Graph support for THD attention.
 * **[07/2026]** `Transformer Engine v2.17 <https://github.com/NVIDIA/TransformerEngine/releases/tag/v2.17>`_ introduces NCCL-backed EP for PyTorch and JAX, faster MoE routing, and a JAX Flax MoE block.
 * **[06/2026]** `Boosting MoE Training Throughput with Advanced Fusion Kernels <https://developer.nvidia.com/blog/boosting-moe-training-throughput-with-advanced-fusion-kernels/>`_ shows how fused Transformer Engine operations accelerate MoE training.

docs/examples/attention/attention.ipynb

@@ -37,7 +37,7 @@
     "  <tr>\n",
     "    <td rowspan=\"3\">PyTorch</td>\n",
     "    <td>cuDNN attention (`FusedAttention`)</td>\n",
-    "    <td rowspan=\"3\"> [transformer_engine.pytorch.attention](https://github.com/NVIDIA/TransformerEngine/blob/main/transformer_engine/pytorch/attention.py)</td>\n",
+    "    <td rowspan=\"3\"> [transformer_engine.pytorch.attention](https://github.com/NVIDIA/TransformerEngine/tree/main/transformer_engine/pytorch/attention)</td>\n",
     "  </tr>\n",
     "  <tr>\n",
     "    <td> flash-attention (`FlashAttention`)</td>\n",
@@ -476,12 +476,12 @@
     "  </tr>\n",
     "</table>\n",
     "\n",
-    "Some example usage of the different layouts can be found at [test_dpa_qkv_layout](https://github.com/NVIDIA/TransformerEngine/blob/main/tests/pytorch/attention/test_attention.py) and [test_dpa_qkv_layout_thd](https://github.com/NVIDIA/TransformerEngine/blob/main/tests/pytorch/attention/test_attention.py). Transformer Engine also provides a utility function [transformer_engine.pytorch.attention.dot_product_attention.utils.get_qkv_layout](https://github.com/NVIDIA/TransformerEngine/blob/main/transformer_engine/pytorch/attention.py) to help determine which layout a set of `q`, `k`, `v` tensors have (PyTorch only).\n",
+    "Some example usage of the different layouts can be found at [test_dpa_qkv_layout](https://github.com/NVIDIA/TransformerEngine/blob/main/tests/pytorch/attention/test_attention.py) and [test_dpa_qkv_layout_thd](https://github.com/NVIDIA/TransformerEngine/blob/main/tests/pytorch/attention/test_attention.py). Transformer Engine also provides a utility function [transformer_engine.pytorch.attention.dot_product_attention.utils.get_qkv_layout](https://github.com/NVIDIA/TransformerEngine/blob/main/transformer_engine/pytorch/attention/dot_product_attention/utils.py) to help determine which layout a set of `q`, `k`, `v` tensors have (PyTorch only).\n",
     "\n",
     "<div class=\"alert alert-info\">\n",
     "<b>Note</b>\n",
     "    \n",
-    "When RoPE is employed, the <code>qkv_layout</code> may change in Transformer Engine PyTorch through [get_qkv_layout](https://github.com/NVIDIA/TransformerEngine/blob/main/transformer_engine/pytorch/attention.py). This is due to the in-place nature of our RoPE implementations. We convert `q`, `k`, `v` tensors from their initial layout to the corresponding <code>hd_hd_hd</code> layout. For example, from <code>sbh3d</code> in <code>pytorch.MultiHeadAttention</code> before RoPE, to <code>sbhd_sbhd_sbhd</code> in <code>pytorch.DotProductAttention</code> after RoPE.\n",
+    "When RoPE is employed, the <code>qkv_layout</code> may change in Transformer Engine PyTorch through [get_qkv_layout](https://github.com/NVIDIA/TransformerEngine/blob/main/transformer_engine/pytorch/attention/dot_product_attention/utils.py). This is due to the in-place nature of our RoPE implementations. We convert `q`, `k`, `v` tensors from their initial layout to the corresponding <code>hd_hd_hd</code> layout. For example, from <code>sbh3d</code> in <code>pytorch.MultiHeadAttention</code> before RoPE, to <code>sbhd_sbhd_sbhd</code> in <code>pytorch.DotProductAttention</code> after RoPE.\n",
     "</div>\n"
    ]
   },

docs/examples/te_jax_integration.rst

@@ -40,7 +40,7 @@ Quantization recipes at a glance
 
 TE exposes its quantization choices as **recipes**. Please see
 `Low-precision Training
-<https://docs.nvidia.com/deeplearning/transformer-engine/user-guide/features/low_precision_training/index.html>`_
+<https://docs.nvidia.com/deeplearning/transformer-engine/features/low_precision_training/index.html>`_
 for a more detailed description of each recipe.
 
 ..  _jax_recipe_table_overview:

docs/examples/te_llama/tutorial_accelerate_hf_llama_with_te.ipynb

@@ -92,7 +92,7 @@
         "\n",
         "<figure align=\"center\">\n",
         "<img src=\"media/transformer_llama.png\">\n",
-        "    <figcaption> Fig 1: Llama visualized as a transformer. (generated with [Nvidia's AI-foundation models](https://catalog.ngc.nvidia.com/orgs/nvidia/teams/ai-foundation/models/sdxl))</figcaption>\n",
+        "    <figcaption> Fig 1: Llama visualized as a transformer. (generated with NVIDIA's AI-foundation models)</figcaption>\n",
         "</figure>\n",
         "\n",
         "A flashback:\n",
@@ -353,7 +353,7 @@
         "\n",
         "<b>Note</b>\n",
         "    \n",
-        "Here, we go over some of the options in `TransformerLayer` that are needed for the tutorial. For a complete list of options, refer the [TransformerLayer API documentation](https://docs.nvidia.com/deeplearning/transformer-engine/user-guide/api/pytorch.html?highlight=transformerlayer#transformer_engine.pytorch.TransformerLayer).\n",
+        "Here, we go over some of the options in `TransformerLayer` that are needed for the tutorial. For a complete list of options, refer the [TransformerLayer API documentation](https://docs.nvidia.com/deeplearning/transformer-engine/api/pytorch.html#transformer_engine.pytorch.TransformerLayer).\n",
         "\n",
         "</div>\n",
         "\n",

docs/examples/te_mixtral/tutorial_accelerate_hf_mixtral_with_te.ipynb

@@ -309,7 +309,7 @@
       "source": [
         "## [Improvement 3] Precision Optimization and Fused MLP (Precision: `MXFP8`)\n",
         "\n",
-        "With EP and grouped expert GEMMs in place, the next improvement lowers precision from `BF16` to `MXFP8`. `MXFP8` converts the weight and activation values to 8 bits instead of 16 bits. To preserve dynamic range, it keeps one `E8M0` scale factor for every 32 values; applying that scale recovers a wider numerical range. On Blackwell GPUs, `MXFP8` is native and hardware accelerated, so `MXFP8` GEMMs can run through specialized Tensor Core instructions. Read more about `MXFP8` and block scaling in the [Transformer Engine FP8 primer](https://docs.nvidia.com/deeplearning/transformer-engine/user-guide/examples/fp8_primer.html#MXFP8-and-block-scaling).\n",
+        "With EP and grouped expert GEMMs in place, the next improvement lowers precision from `BF16` to `MXFP8`. `MXFP8` converts the weight and activation values to 8 bits instead of 16 bits. To preserve dynamic range, it keeps one `E8M0` scale factor for every 32 values; applying that scale recovers a wider numerical range. On Blackwell GPUs, `MXFP8` is native and hardware accelerated, so `MXFP8` GEMMs can run through specialized Tensor Core instructions. Read more about `MXFP8` and block scaling in the [Transformer Engine FP8 primer](https://docs.nvidia.com/deeplearning/transformer-engine/examples/fp8_primer.html#MXFP8-and-block-scaling).\n",
         "\n",
         "The model still keeps its master weights in `BF16`, so using `MXFP8` adds `Quantization` and `De-Quantization` work around the GEMMs. `Quantization` converts `BF16` weights and activations to `MXFP8` before the low-precision GEMM; `De-Quantization` converts the result back to the higher-precision format. The naive path performs these as separate operations. This motivates the fused MLP path shown below.\n",
         "\n",

docs/features/low_precision_training/introduction/introduction.rst

@@ -231,7 +231,7 @@ Let's now see how we can train in lower precisions in supported frameworks.
       .. note::
          Python context managers like ``autocast`` may interact unexpectedly with JAX's JIT compilation.
          For finer-grained control, consider passing the recipe directly to TE modules instead.
-         See the `TE JAX Integration notebook <https://github.com/NVIDIA/TransformerEngine/blob/main/docs/examples/te_jax_integration.ipynb>`_
+         See the `TE JAX Integration guide <https://github.com/NVIDIA/TransformerEngine/blob/main/docs/examples/te_jax_integration.rst>`_
          for details.
 
 **Mixed precision with 8- or 4-bit precisions**

docs/index.rst

@@ -12,7 +12,7 @@ Transformer Engine documentation
       You are currently viewing unstable developer preview of the documentation.
       To see the documentation for the latest stable release, refer to:
 
-      * `Release Notes <https://docs.nvidia.com/deeplearning/transformer-engine/release-notes/index.html>`_
+      * `Release Notes <https://docs.nvidia.com/deeplearning/transformer-engine/release_notes.html>`_
       * `Developer Guide <https://docs.nvidia.com/deeplearning/transformer-engine/index.html>`_ (stable version of this page)
 
 .. include:: ../README.rst

docs/project_updates.rst

@@ -13,7 +13,7 @@ The README highlights only the most recent updates.
 ----
 
 * **[09/2026]** `Transformer Engine v2.19 <https://github.com/NVIDIA/TransformerEngine/releases/tag/v2.19>`_ adds Rubin support, hybrid quantization, MXFP8 EP communication, and expanded FP8 attention support.
-* **[09/2026]** `Accelerating Dropless MoE Training in JAX with NVIDIA Transformer Engine <https://developer.nvidia.com/blog/accelerating-dropless-moe-training-in-jax-with-nvidia-transform-engine/>`_.
+* **[09/2026]** `Accelerating Dropless MoE Training in JAX with NVIDIA Transformer Engine <https://developer.nvidia.com/blog/accelerating-dropless-moe-training-in-jax-with-nvidia-transformer-engine/>`_.
 * **[08/2026]** `Transformer Engine v2.18 <https://github.com/NVIDIA/TransformerEngine/releases/tag/v2.18>`_.
 * **[07/2026]** `Transformer Engine v2.17 <https://github.com/NVIDIA/TransformerEngine/releases/tag/v2.17>`_.
 * **[06/2026]** `Boosting MoE Training Throughput with Advanced Fusion Kernels <https://developer.nvidia.com/blog/boosting-moe-training-throughput-with-advanced-fusion-kernels/>`_.
@@ -53,7 +53,7 @@ The README highlights only the most recent updates.
 * **[03/2024]** `Turbocharged Training: Optimizing the Databricks Mosaic AI Stack with FP8 <https://www.databricks.com/blog/turbocharged-training-optimizing-databricks-mosaic-ai-stack-fp8>`_.
 * **[03/2024]** `FP8 Training Support in SageMaker Model Parallelism Library <https://docs.aws.amazon.com/sagemaker/latest/dg/model-parallel-release-notes.html>`_.
 * **[12/2023]** `New NVIDIA NeMo Framework Features and NVIDIA H200 <https://developer.nvidia.com/blog/new-nvidia-nemo-framework-features-and-nvidia-h200-supercharge-llm-training-performance-and-versatility/>`_.
-* **[11/2023]** `Inflection-2: The Next Step Up <https://inflection.ai/inflection-2>`_.
+* **[11/2023]** Inflection-2: The Next Step Up.
 * **[11/2023]** `Unleashing the Power of Transformers with NVIDIA Transformer Engine <https://lambda.ai/blog/unleashing-the-power-of-transformers-with-nvidia-transformer-engine>`_.
 * **[11/2023]** `Accelerating PyTorch Training Workloads with FP8 <https://towardsdatascience.com/accelerating-pytorch-training-workloads-with-fp8-5a5123aec7d7>`_.
 * **[09/2023]** `Transformer Engine Added to the AWS Deep Learning Container for PyTorch Training <https://github.com/aws/deep-learning-containers/pull/3315>`_.

examples/README.md

@@ -32,7 +32,7 @@ Additionally, we offer [Jupyter notebook tutorials](https://github.com/NVIDIA/Tr
   - Model Parallelism: Divide a model across multiple GPUs for parallel training.
   - Multiprocessing with Model Parallelism: Multiprocessing for model parallelism, including multi-node support and hardware affinity setup.
 - [Basic MNIST Example](https://github.com/NVIDIA/TransformerEngine/tree/main/examples/jax/mnist)
-- [TE JAX Integration Tutorial](https://github.com/NVIDIA/TransformerEngine/blob/main/docs/examples/te_jax_integration.ipynb)
+- [TE JAX Integration Guide](https://github.com/NVIDIA/TransformerEngine/blob/main/docs/examples/te_jax_integration.rst)
   - Introduction to integrating TE into an existing JAX model framework, building a Transformer Layer, and instructions on integrating TE modules like Linear and LayerNorm.
 
 # Third party

examples/jax/encoder/README.md

@@ -42,7 +42,7 @@ python test_single_gpu_encoder.py --use-fp8
    export CUDA_VISIBLE_DEVICES=0,1,2,3
    python test_multigpu_encoder.py
    ```
-   Please refer to [CUDA Environment Variables](https://docs.nvidia.com/cuda/cuda-c-programming-guide/#cuda-environment-variables) for more details.
+   Please refer to [CUDA Environment Variables](https://docs.nvidia.com/cuda/cuda-programming-guide/05-appendices/environment-variables.html) for more details.
 
 ### Run ###
 
@@ -100,7 +100,7 @@ python test_model_parallel_encoder.py --use-fp8
    GPU7    80-95,208-223   5
    ```
 
-4. It is recommended to set the environment variable `CUDA_DEVICE_ORDER` to `PCI_BUS_ID` before running the example with the affinity setting. To ensure that the device order is aligned between CUDA and `nvidia-smi`. Please refer to [CUDA Environment Variables](https://docs.nvidia.com/cuda/cuda-c-programming-guide/#cuda-environment-variables) for more details.
+4. It is recommended to set the environment variable `CUDA_DEVICE_ORDER` to `PCI_BUS_ID` before running the example with the affinity setting. To ensure that the device order is aligned between CUDA and `nvidia-smi`. Please refer to [CUDA Environment Variables](https://docs.nvidia.com/cuda/cuda-programming-guide/05-appendices/environment-variables.html) for more details.
 
 5. `jax.distributed.initialize` must be called before any other JAX or Flax API, otherwise `jax.local_devices` will be incorrect. `jax.distributed.shutdown` should be the last API call.