Reflect RoCm support in the documentation (#27636)

* reflect RoCm support in the documentation

* Update docs/source/en/main_classes/trainer.md
Co-authored-by: Lysandre Debut <hi@lysand.re>

* fix review comments

* use ROCm instead of RoCm

---------
Co-authored-by: Lysandre Debut <hi@lysand.re>

docs/source/en/main_classes/trainer.md

@@ -26,7 +26,7 @@ If you're looking to fine-tune a language model like Llama-2 or Mistral on a tex
 
 Before instantiating your [`Trainer`], create a [`TrainingArguments`] to access all the points of customization during training.
 
-The API supports distributed training on multiple GPUs/TPUs, mixed precision through [NVIDIA Apex](https://github.com/NVIDIA/apex) and Native AMP for PyTorch.
+The API supports distributed training on multiple GPUs/TPUs, mixed precision through [NVIDIA Apex] for NVIDIA GPUs, [ROCm APEX](https://github.com/ROCmSoftwarePlatform/apex) for AMD GPUs, and Native AMP for PyTorch.
 
 The [`Trainer`] contains the basic training loop which supports the above features. To inject custom behavior you can subclass them and override the following methods:
 
@@ -272,7 +272,7 @@ but this approach can be confusing since you may forget you set up the environme
 
 There is an additional environment variable `CUDA_DEVICE_ORDER` that controls how the physical devices are ordered. The two choices are:
 
-1. ordered by PCIe bus IDs (matches `nvidia-smi`'s order) - this is the default.
+1. ordered by PCIe bus IDs (matches `nvidia-smi` and `rocm-smi`'s order) - this is the default.
 
 ```bash
 export CUDA_DEVICE_ORDER=PCI_BUS_ID
@@ -284,7 +284,7 @@ export CUDA_DEVICE_ORDER=PCI_BUS_ID
 export CUDA_DEVICE_ORDER=FASTEST_FIRST
 ```
 
-Most of the time you don't need to care about this environment variable, but it's very helpful if you have a lopsided setup where you have an old and a new GPUs physically inserted in such a way so that the slow older card appears to be first. One way to fix that is to swap the cards. But if you can't swap the cards (e.g., if the cooling of the devices gets impacted) then setting `CUDA_DEVICE_ORDER=FASTEST_FIRST` will always put the newer faster card first. It'll be somewhat confusing though since `nvidia-smi` will still report them in the PCIe order.
+Most of the time you don't need to care about this environment variable, but it's very helpful if you have a lopsided setup where you have an old and a new GPUs physically inserted in such a way so that the slow older card appears to be first. One way to fix that is to swap the cards. But if you can't swap the cards (e.g., if the cooling of the devices gets impacted) then setting `CUDA_DEVICE_ORDER=FASTEST_FIRST` will always put the newer faster card first. It'll be somewhat confusing though since `nvidia-smi` (or `rocm-smi`) will still report them in the PCIe order.
 
 The other solution to swapping the order is to use:
 

docs/source/en/model_doc/t5.md

@@ -314,7 +314,7 @@ The predicted tokens will then be placed between the sentinel tokens.
 
 ## Performance
 
-If you'd like a faster training and inference performance, install [apex](https://github.com/NVIDIA/apex#quick-start) and then the model will automatically use `apex.normalization.FusedRMSNorm` instead of `T5LayerNorm`. The former uses an optimized fused kernel which is several times faster than the latter.
+If you'd like a faster training and inference performance, install [NVIDIA APEX](https://github.com/NVIDIA/apex#quick-start) for NVIDIA GPUs, or [ROCm APEX](https://github.com/ROCmSoftwarePlatform/apex) for AMD GPUs and then the model will automatically use `apex.normalization.FusedRMSNorm` instead of `T5LayerNorm`. The former uses an optimized fused kernel which is several times faster than the latter.
 
 
 ## Resources

docs/source/en/perf_infer_gpu_one.md

@@ -15,7 +15,7 @@ rendered properly in your Markdown viewer.
 
 # GPU inference
 
-GPUs are the standard choice of hardware for machine learning, unlike CPUs, because they are optimized for memory bandwidth and parallelism. To keep up with the larger sizes of modern models or to run these large models on existing and older hardware, there are several optimizations you can use to speed up GPU inference. In this guide, you'll learn how to use FlashAttention-2 (a more memory-efficient attention mechanism), BetterTransformer (a PyTorch native fastpath execution), and bitsandbytes to quantize your model to a lower precision. Finally, learn how to use 🤗 Optimum to accelerate inference with ONNX Runtime on Nvidia GPUs.
+GPUs are the standard choice of hardware for machine learning, unlike CPUs, because they are optimized for memory bandwidth and parallelism. To keep up with the larger sizes of modern models or to run these large models on existing and older hardware, there are several optimizations you can use to speed up GPU inference. In this guide, you'll learn how to use FlashAttention-2 (a more memory-efficient attention mechanism), BetterTransformer (a PyTorch native fastpath execution), and bitsandbytes to quantize your model to a lower precision. Finally, learn how to use 🤗 Optimum to accelerate inference with ONNX Runtime on Nvidia and AMD GPUs.
 
 <Tip>
 
@@ -276,13 +276,13 @@ Feel free to try running a 11 billion parameter [T5 model](https://colab.researc
 
 <Tip>
 
-Learn more details about using ORT with 🤗 Optimum in the [Accelerated inference on NVIDIA GPUs](https://huggingface.co/docs/optimum/onnxruntime/usage_guides/gpu#accelerated-inference-on-nvidia-gpus) guide. This section only provides a brief and simple example.
+Learn more details about using ORT with 🤗 Optimum in the [Accelerated inference on NVIDIA GPUs](https://huggingface.co/docs/optimum/onnxruntime/usage_guides/gpu#accelerated-inference-on-nvidia-gpus) and [Accelerated inference on AMD GPUs](https://huggingface.co/docs/optimum/onnxruntime/usage_guides/amdgpu#accelerated-inference-on-amd-gpus) guides. This section only provides a brief and simple example.
 
 </Tip>
 
-ONNX Runtime (ORT) is a model accelerator that supports accelerated inference on Nvidia GPUs. ORT uses optimization techniques like fusing common operations into a single node and constant folding to reduce the number of computations performed and speedup inference. ORT also places the most computationally intensive operations on the GPU and the rest on the CPU to intelligently distribute the workload between the two devices.
+ONNX Runtime (ORT) is a model accelerator that supports accelerated inference on Nvidia GPUs, and AMD GPUs that use [ROCm](https://www.amd.com/en/products/software/rocm.html) stack. ORT uses optimization techniques like fusing common operations into a single node and constant folding to reduce the number of computations performed and speedup inference. ORT also places the most computationally intensive operations on the GPU and the rest on the CPU to intelligently distribute the workload between the two devices.
 
-ORT is supported by 🤗 Optimum which can be used in 🤗 Transformers. You'll need to use an [`~optimum.onnxruntime.ORTModel`] for the task you're solving, and specify the `provider` parameter which can be set to either [`CUDAExecutionProvider`](https://huggingface.co/docs/optimum/onnxruntime/usage_guides/gpu#cudaexecutionprovider) or [`TensorrtExecutionProvider`](https://huggingface.co/docs/optimum/onnxruntime/usage_guides/gpu#tensorrtexecutionprovider). If you want to load a model that was not yet exported to ONNX, you can set `export=True` to convert your model on-the-fly to the ONNX format :
+ORT is supported by 🤗 Optimum which can be used in 🤗 Transformers. You'll need to use an [`~optimum.onnxruntime.ORTModel`] for the task you're solving, and specify the `provider` parameter which can be set to either [`CUDAExecutionProvider`](https://huggingface.co/docs/optimum/onnxruntime/usage_guides/gpu#cudaexecutionprovider), [`ROCMExecutionProvider`](https://huggingface.co/docs/optimum/onnxruntime/usage_guides/amdgpu) or [`TensorrtExecutionProvider`](https://huggingface.co/docs/optimum/onnxruntime/usage_guides/gpu#tensorrtexecutionprovider). If you want to load a model that was not yet exported to ONNX, you can set `export=True` to convert your model on-the-fly to the ONNX format:
 
 ```py
 from optimum.onnxruntime import ORTModelForSequenceClassification

docs/source/en/perf_train_gpu_one.md

@@ -237,7 +237,7 @@ You can speedup the training throughput by using Flash Attention 2 integration i
 The most common optimizer used to train transformer models is Adam or AdamW (Adam with weight decay). Adam achieves 
 good convergence by storing the rolling average of the previous gradients; however, it adds an additional memory 
 footprint of the order of the number of model parameters. To remedy this, you can use an alternative optimizer. 
-For example if you have [NVIDIA/apex](https://github.com/NVIDIA/apex) installed, `adamw_apex_fused` will give you the 
+For example if you have [NVIDIA/apex](https://github.com/NVIDIA/apex) installed for NVIDIA GPUs, or [ROCmSoftwarePlatform/apex](https://github.com/ROCmSoftwarePlatform/apex) for AMD GPUs, `adamw_apex_fused` will give you the
 fastest training experience among all supported AdamW optimizers.
 
 [`Trainer`] integrates a variety of optimizers that can be used out of box: `adamw_hf`, `adamw_torch`, `adamw_torch_fused`, 

docs/source/en/tasks/text-to-speech.md

@@ -74,6 +74,12 @@ To follow this guide you will need a GPU. If you're working in a notebook, run t
 !nvidia-smi
 ```
 
+or alternatively for AMD GPUs:
+
+```bash
+!rocm-smi
+```
+
 </Tip>
 
 We encourage you to log in to your Hugging Face account to upload and share your model with the community. When prompted, enter your token to log in: