[docs] fixed links with 404 (#27327)

* fixed links with 404

* make style

docs/source/en/create_a_model.md

@@ -110,7 +110,7 @@ You can also save your configuration file as a dictionary or even just the diffe
 
 ## Model
 
-The next step is to create a [model](main_classes/models). The model - also loosely referred to as the architecture - defines what each layer is doing and what operations are happening. Attributes like `num_hidden_layers` from the configuration are used to define the architecture. Every model shares the base class [`PreTrainedModel`] and a few common methods like resizing input embeddings and pruning self-attention heads. In addition, all models are also either a [`torch.nn.Module`](https://pytorch.org/docs/stable/generated/torch.nn.Module.html), [`tf.keras.Model`](https://www.tensorflow.org/api_docs/python/tf/keras/Model) or [`flax.linen.Module`](https://flax.readthedocs.io/en/latest/flax.linen.html#module) subclass. This means models are compatible with each of their respective framework's usage.
+The next step is to create a [model](main_classes/models). The model - also loosely referred to as the architecture - defines what each layer is doing and what operations are happening. Attributes like `num_hidden_layers` from the configuration are used to define the architecture. Every model shares the base class [`PreTrainedModel`] and a few common methods like resizing input embeddings and pruning self-attention heads. In addition, all models are also either a [`torch.nn.Module`](https://pytorch.org/docs/stable/generated/torch.nn.Module.html), [`tf.keras.Model`](https://www.tensorflow.org/api_docs/python/tf/keras/Model) or [`flax.linen.Module`](https://flax.readthedocs.io/en/latest/api_reference/flax.linen/module.html) subclass. This means models are compatible with each of their respective framework's usage.
 
 <frameworkcontent>
 <pt>

docs/source/en/main_classes/processors.md

@@ -86,7 +86,7 @@ This library hosts the processor to load the XNLI data:
 
 Please note that since the gold labels are available on the test set, evaluation is performed on the test set.
 
-An example using these processors is given in the [run_xnli.py](https://github.com/huggingface/transformers/tree/main/examples/legacy/text-classification/run_xnli.py) script.
+An example using these processors is given in the [run_xnli.py](https://github.com/huggingface/transformers/tree/main/examples/pytorch/text-classification/run_xnli.py) script.
 
 
 ## SQuAD

docs/source/en/main_classes/quantization.md

@@ -95,7 +95,7 @@ The benchmark was run on a NVIDIA-A100 instance and the model used was [`TheBlok
 <img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/quantization/forward_latency_plot.png">
 </div>
 
-You can find the full results together with packages versions in [this link](https://github.com/huggingface/optimum-benchmark/tree/main/examples/running-mistral).
+You can find the full results together with packages versions in [this link](https://github.com/huggingface/optimum-benchmark/tree/main/examples/running-mistrals).
 
 From the results it appears that AWQ quantization method is the fastest quantization method for inference, text generation and among the lowest peak memory for text generation. However, AWQ seems to have the largest forward latency per batch size. 
 

docs/source/en/model_doc/clap.md

@@ -27,7 +27,7 @@ The abstract from the paper is the following:
 
 *Contrastive learning has shown remarkable success in the field of multimodal representation learning. In this paper, we propose a pipeline of contrastive language-audio pretraining to develop an audio representation by combining audio data with natural language descriptions. To accomplish this target, we first release LAION-Audio-630K, a large collection of 633,526 audio-text pairs from different data sources. Second, we construct a contrastive language-audio pretraining model by considering different audio encoders and text encoders. We incorporate the feature fusion mechanism and keyword-to-caption augmentation into the model design to further enable the model to process audio inputs of variable lengths and enhance the performance. Third, we perform comprehensive experiments to evaluate our model across three tasks: text-to-audio retrieval, zero-shot audio classification, and supervised audio classification. The results demonstrate that our model achieves superior performance in text-to-audio retrieval task. In audio classification tasks, the model achieves state-of-the-art performance in the zeroshot setting and is able to obtain performance comparable to models' results in the non-zero-shot setting. LAION-Audio-6*
 
-This model was contributed by [Younes Belkada](https://huggingface.co/ybelkada) and [Arthur Zucker](https://huggingface.co/ArtZucker) .
+This model was contributed by [Younes Belkada](https://huggingface.co/ybelkada) and [Arthur Zucker](https://huggingface.co/ArthurZ) .
 The original code can be found [here](https://github.com/LAION-AI/Clap).
 
 ## ClapConfig

docs/source/en/model_doc/mobilebert.md

@@ -37,7 +37,7 @@ natural language inference tasks of GLUE, MobileBERT achieves a GLUEscore o 77.7
 latency on a Pixel 4 phone. On the SQuAD v1.1/v2.0 question answering task, MobileBERT achieves a dev F1 score of
 90.0/79.2 (1.5/2.1 higher than BERT_BASE).*
 
-This model was contributed by [vshampor](https://huggingface.co/vshampor). The original code can be found [here](https://github.com/google-research/mobilebert).
+This model was contributed by [vshampor](https://huggingface.co/vshampor). The original code can be found [here](https://github.com/google-research/google-research/tree/master/mobilebert).
 
 ## Usage tips
 

docs/source/en/model_doc/nllb-moe.md

@@ -37,7 +37,7 @@ improvements to counteract overfitting while training on thousands of tasks. Cri
 a human-translated benchmark, Flores-200, and combined human evaluation with a novel toxicity benchmark covering all languages in Flores-200 to assess translation safety.
 Our model achieves an improvement of 44% BLEU relative to the previous state-of-the-art, laying important groundwork towards realizing a universal translation system.*
 
-This model was contributed by [Arthur Zucker](https://huggingface.co/ArtZucker).
+This model was contributed by [Arthur Zucker](https://huggingface.co/ArthurZ).
 The original code can be found [here](https://github.com/facebookresearch/fairseq).
 
 ## Usage tips

docs/source/en/model_doc/regnet.md

@@ -27,7 +27,7 @@ The abstract from the paper is the following:
 *In this work, we present a new network design paradigm. Our goal is to help advance the understanding of network design and discover design principles that generalize across settings. Instead of focusing on designing individual network instances, we design network design spaces that parametrize populations of networks. The overall process is analogous to classic manual design of networks, but elevated to the design space level. Using our methodology we explore the structure aspect of network design and arrive at a low-dimensional design space consisting of simple, regular networks that we call RegNet. The core insight of the RegNet parametrization is surprisingly simple: widths and depths of good networks can be explained by a quantized linear function. We analyze the RegNet design space and arrive at interesting findings that do not match the current practice of network design. The RegNet design space provides simple and fast networks that work well across a wide range of flop regimes. Under comparable training settings and flops, the RegNet models outperform the popular EfficientNet models while being up to 5x faster on GPUs.*
 
 This model was contributed by [Francesco](https://huggingface.co/Francesco). The TensorFlow version of the model
-was contributed by [sayakpaul](https://huggingface.com/sayakpaul) and [ariG23498](https://huggingface.com/ariG23498).
+was contributed by [sayakpaul](https://huggingface.co/sayakpaul) and [ariG23498](https://huggingface.co/ariG23498).
 The original code can be found [here](https://github.com/facebookresearch/pycls).
 
 The huge 10B model from [Self-supervised Pretraining of Visual Features in the Wild](https://arxiv.org/abs/2103.01988), 

docs/source/en/model_doc/roberta-prelayernorm.md

@@ -25,7 +25,7 @@ The abstract from the paper is the following:
 
 *fairseq is an open-source sequence modeling toolkit that allows researchers and developers to train custom models for translation, summarization, language modeling, and other text generation tasks. The toolkit is based on PyTorch and supports distributed training across multiple GPUs and machines. We also support fast mixed-precision training and inference on modern GPUs.*
 
-This model was contributed by [andreasmaden](https://huggingface.co/andreasmaden).
+This model was contributed by [andreasmaden](https://huggingface.co/andreasmadsen).
 The original code can be found [here](https://github.com/princeton-nlp/DinkyTrain).
 
 ## Usage tips

docs/source/en/model_doc/switch_transformers.md

@@ -27,7 +27,7 @@ The abstract from the paper is the following:
 
 *In deep learning, models typically reuse the same parameters for all inputs. Mixture of Experts (MoE) defies this and instead selects different parameters for each incoming example. The result is a sparsely-activated model -- with outrageous numbers of parameters -- but a constant computational cost. However, despite several notable successes of MoE, widespread adoption has been hindered by complexity, communication costs and training instability -- we address these with the Switch Transformer. We simplify the MoE routing algorithm and design intuitive improved models with reduced communication and computational costs. Our proposed training techniques help wrangle the instabilities and we show large sparse models may be trained, for the first time, with lower precision (bfloat16) formats. We design models based off T5-Base and T5-Large to obtain up to 7x increases in pre-training speed with the same computational resources. These improvements extend into multilingual settings where we measure gains over the mT5-Base version across all 101 languages. Finally, we advance the current scale of language models by pre-training up to trillion parameter models on the "Colossal Clean Crawled Corpus" and achieve a 4x speedup over the T5-XXL model.*
 
-This model was contributed by [Younes Belkada](https://huggingface.co/ybelkada) and [Arthur Zucker](https://huggingface.co/ArtZucker) .
+This model was contributed by [Younes Belkada](https://huggingface.co/ybelkada) and [Arthur Zucker](https://huggingface.co/ArthurZ).
 The original code can be found [here](https://github.com/google/flaxformer/tree/main/flaxformer/architectures/moe).
 
 ## Usage tips

docs/source/en/model_doc/tapas.md

@@ -47,7 +47,7 @@ This model was contributed by [nielsr](https://huggingface.co/nielsr). The Tenso
 ## Usage tips
 
 - TAPAS is a model that uses relative position embeddings by default (restarting the position embeddings at every cell of the table). Note that this is something that was added after the publication of the original TAPAS paper. According to the authors, this usually results in a slightly better performance, and allows you to encode longer sequences without running out of embeddings. This is reflected in the `reset_position_index_per_cell` parameter of [`TapasConfig`], which is set to `True` by default. The default versions of the models available on the [hub](https://huggingface.co/models?search=tapas) all use relative position embeddings. You can still use the ones with absolute position embeddings by passing in an additional argument `revision="no_reset"` when calling the `from_pretrained()` method. Note that it's usually advised to pad the inputs on the right rather than the left.
-- TAPAS is based on BERT, so `TAPAS-base` for example corresponds to a `BERT-base` architecture. Of course, `TAPAS-large` will result in the best performance (the results reported in the paper are from `TAPAS-large`). Results of the various sized models are shown on the [original Github repository](https://github.com/google-research/tapas>).
+- TAPAS is based on BERT, so `TAPAS-base` for example corresponds to a `BERT-base` architecture. Of course, `TAPAS-large` will result in the best performance (the results reported in the paper are from `TAPAS-large`). Results of the various sized models are shown on the [original GitHub repository](https://github.com/google-research/tapas).
 - TAPAS has checkpoints fine-tuned on SQA, which are capable of answering questions related to a table in a conversational set-up. This means that you can ask follow-up questions such as "what is his age?" related to the previous question. Note that the forward pass of TAPAS is a bit different in case of a conversational set-up: in that case, you have to feed every table-question pair one by one to the model, such that the `prev_labels` token type ids can be overwritten by the predicted `labels` of the model to the previous question. See "Usage" section for more info.
 - TAPAS is similar to BERT and therefore relies on the masked language modeling (MLM) objective. It is therefore efficient at predicting masked tokens and at NLU in general, but is not optimal for text generation. Models trained with a causal language modeling (CLM) objective are better in that regard. Note that TAPAS can be used as an encoder in the EncoderDecoderModel framework, to combine it with an autoregressive text decoder such as GPT-2.
 

docs/source/en/perf_train_gpu_many.md

@@ -270,7 +270,7 @@ which is discussed next.
 
 Implementations:
 
-- [DeepSpeed](https://www.deepspeed.ai/features/#the-zero-redundancy-optimizer) ZeRO-DP stages 1+2+3
+- [DeepSpeed](https://www.deepspeed.ai/tutorials/zero/) ZeRO-DP stages 1+2+3
 - [`Accelerate` integration](https://huggingface.co/docs/accelerate/en/usage_guides/deepspeed) 
 - [`transformers` integration](main_classes/trainer#trainer-integrations)
 
@@ -434,7 +434,7 @@ This section is based on the original much more [detailed TP overview](https://g
 by [@anton-l](https://github.com/anton-l).
 
 Alternative names:
-- DeepSpeed calls it [tensor slicing](https://www.deepspeed.ai/features/#model-parallelism)
+- DeepSpeed calls it [tensor slicing](https://www.deepspeed.ai/training/#model-parallelism)
 
 Implementations:
 - [Megatron-LM](https://github.com/NVIDIA/Megatron-LM) has an internal implementation, as it's very model-specific

docs/source/en/perf_train_gpu_one.md

@@ -394,7 +394,7 @@ Choose which backend to use by specifying it via `torch_compile_backend` in the
 
 **Inference-only backend**s:
 * `dynamo.optimize("ofi")` -  Uses Torchscript optimize_for_inference.  [Read more](https://pytorch.org/docs/stable/generated/torch.jit.optimize_for_inference.html)
-* `dynamo.optimize("fx2trt")` -  Uses Nvidia TensorRT for inference optimizations.  [Read more](https://github.com/pytorch/TensorRT/blob/master/docsrc/tutorials/getting_started_with_fx_path.rst)
+* `dynamo.optimize("fx2trt")` -  Uses NVIDIA TensorRT for inference optimizations.  [Read more](https://pytorch.org/TensorRT/tutorials/getting_started_with_fx_path.html)
 * `dynamo.optimize("onnxrt")` -  Uses ONNXRT for inference on CPU/GPU.  [Read more](https://onnxruntime.ai/)
 * `dynamo.optimize("ipex")` -  Uses IPEX for inference on CPU.  [Read more](https://github.com/intel/intel-extension-for-pytorch)
 
@@ -505,7 +505,7 @@ Most related papers and implementations are built around Tensorflow/TPUs:
 - [Switch Transformers: Scaling to Trillion Parameter Models with Simple and Efficient Sparsity](https://arxiv.org/abs/2101.03961)
 - [GLaM: Generalist Language Model (GLaM)](https://ai.googleblog.com/2021/12/more-efficient-in-context-learning-with.html)
 
-And for Pytorch DeepSpeed has built one as well: [DeepSpeed-MoE: Advancing Mixture-of-Experts Inference and Training to Power Next-Generation AI Scale](https://arxiv.org/abs/2201.05596), [Mixture of Experts](https://www.deepspeed.ai/tutorials/mixture-of-experts/) - blog posts:  [1](https://www.microsoft.com/en-us/research/blog/deepspeed-powers-8x-larger-moe-model-training-with-high-performance/), [2](https://www.microsoft.com/en-us/research/publication/scalable-and-efficient-moe-training-for-multitask-multilingual-models/) and specific deployment with large transformer-based natural language generation models: [blog post](https://www.deepspeed.ai/news/2021/12/09/deepspeed-moe-nlg.html), [Megatron-Deepspeed branch](Thttps://github.com/microsoft/Megatron-DeepSpeed/tree/moe-training).
+And for Pytorch DeepSpeed has built one as well: [DeepSpeed-MoE: Advancing Mixture-of-Experts Inference and Training to Power Next-Generation AI Scale](https://arxiv.org/abs/2201.05596), [Mixture of Experts](https://www.deepspeed.ai/tutorials/mixture-of-experts/) - blog posts:  [1](https://www.microsoft.com/en-us/research/blog/deepspeed-powers-8x-larger-moe-model-training-with-high-performance/), [2](https://www.microsoft.com/en-us/research/publication/scalable-and-efficient-moe-training-for-multitask-multilingual-models/) and specific deployment with large transformer-based natural language generation models: [blog post](https://www.deepspeed.ai/2021/12/09/deepspeed-moe-nlg.html), [Megatron-Deepspeed branch](https://github.com/microsoft/Megatron-DeepSpeed/tree/moe-training).
 
 ## Using PyTorch native attention and Flash Attention
 

docs/source/en/philosophy.md

@@ -64,7 +64,7 @@ A few other goals:
 
 The library is built around three types of classes for each model:
 
-- **Model classes** can be PyTorch models ([torch.nn.Module](https://pytorch.org/docs/stable/nn.html#torch.nn.Module)), Keras models ([tf.keras.Model](https://www.tensorflow.org/api_docs/python/tf/keras/Model)) or JAX/Flax models ([flax.linen.Module](https://flax.readthedocs.io/en/latest/api_reference/flax.linen.html)) that work with the pretrained weights provided in the library.
+- **Model classes** can be PyTorch models ([torch.nn.Module](https://pytorch.org/docs/stable/nn.html#torch.nn.Module)), Keras models ([tf.keras.Model](https://www.tensorflow.org/api_docs/python/tf/keras/Model)) or JAX/Flax models ([flax.linen.Module](https://flax.readthedocs.io/en/latest/api_reference/flax.linen/module.html)) that work with the pretrained weights provided in the library.
 - **Configuration classes** store the hyperparameters required to build a model (such as the number of layers and hidden size). You don't always need to instantiate these yourself. In particular, if you are using a pretrained model without any modification, creating the model will automatically take care of instantiating the configuration (which is part of the model).
 - **Preprocessing classes** convert the raw data into a format accepted by the model. A [tokenizer](main_classes/tokenizer) stores the vocabulary for each model and provide methods for encoding and decoding strings in a list of token embedding indices to be fed to a model. [Image processors](main_classes/image_processor) preprocess vision inputs, [feature extractors](main_classes/feature_extractor) preprocess audio inputs, and a [processor](main_classes/processors) handles multimodal inputs.
 

docs/source/en/sagemaker.md

@@ -26,4 +26,3 @@ The documentation has been moved to [hf.co/docs/sagemaker](https://huggingface.c
 
 - [Train Hugging Face models on Amazon SageMaker with the SageMaker Python SDK](https://huggingface.co/docs/sagemaker/train)
 - [Deploy Hugging Face models to Amazon SageMaker with the SageMaker Python SDK](https://huggingface.co/docs/sagemaker/inference)
-- [Frequently Asked Questions](https://huggingface.co/docs/sagemaker/faq)

docs/source/es/create_a_model.md

@@ -109,7 +109,7 @@ También puedes guardar los archivos de configuración como un diccionario; o in
 
 ## Modelo
 
-El siguiente paso será crear un [modelo](main_classes/models). El modelo, al que a veces también nos referimos como arquitectura, es el encargado de definir cada capa y qué operaciones se realizan. Los atributos como `num_hidden_layers` de la configuración se usan para definir la arquitectura. Todos los modelos comparten una clase base, [`PreTrainedModel`], y algunos métodos comunes que se pueden usar para redimensionar los _embeddings_ o para recortar cabezas de auto-atención (también llamadas _self-attention heads_). Además, todos los modelos son subclases de [`torch.nn.Module`](https://pytorch.org/docs/stable/generated/torch.nn.Module.html), [`tf.keras.Model`](https://www.tensorflow.org/api_docs/python/tf/keras/Model) o [`flax.linen.Module`](https://flax.readthedocs.io/en/latest/flax.linen.html#module), lo que significa que son compatibles con su respectivo framework. 
+El siguiente paso será crear un [modelo](main_classes/models). El modelo, al que a veces también nos referimos como arquitectura, es el encargado de definir cada capa y qué operaciones se realizan. Los atributos como `num_hidden_layers` de la configuración se usan para definir la arquitectura. Todos los modelos comparten una clase base, [`PreTrainedModel`], y algunos métodos comunes que se pueden usar para redimensionar los _embeddings_ o para recortar cabezas de auto-atención (también llamadas _self-attention heads_). Además, todos los modelos son subclases de [`torch.nn.Module`](https://pytorch.org/docs/stable/generated/torch.nn.Module.html), [`tf.keras.Model`](https://www.tensorflow.org/api_docs/python/tf/keras/Model) o [`flax.linen.Module`](https://flax.readthedocs.io/en/latest/api_reference/flax.linen/module.html), lo que significa que son compatibles con su respectivo framework. 
 
 <frameworkcontent>
 <pt>

docs/source/es/sagemaker.md

@@ -26,4 +26,3 @@ La documentación ha sido trasladada a [hf.co/docs/sagemaker](https://huggingfac
 
 - [Entrenar modelos de Hugging Face en Amazon SageMaker con SageMaker Python SDK](https://huggingface.co/docs/sagemaker/train)
 - [Desplegar modelos de Hugging Face en Amazon SageMaker con SageMaker Python SDK](https://huggingface.co/docs/sagemaker/inference)
-- [Preguntas Frecuentes](https://huggingface.co/docs/sagemaker/faq)

docs/source/it/create_a_model.md

@@ -109,7 +109,7 @@ Puoi anche salvare il file di configurazione come dizionario oppure come la diff
 
 ## Modello
 
-Il prossimo passo e di creare [modello](main_classes/models). Il modello - vagamente riferito anche come architettura - definisce cosa ogni strato deve fare e quali operazioni stanno succedendo. Attributi come `num_hidden_layers` provenienti dalla configurazione sono usati per definire l'architettura. Ogni modello condivide la classe base [`PreTrainedModel`] e alcuni metodi comuni come il ridimensionamento degli input embeddings e la soppressione delle self-attention heads . Inoltre, tutti i modelli sono la sottoclasse di [`torch.nn.Module`](https://pytorch.org/docs/stable/generated/torch.nn.Module.html), [`tf.keras.Model`](https://www.tensorflow.org/api_docs/python/tf/keras/Model) o [`flax.linen.Module`](https://flax.readthedocs.io/en/latest/flax.linen.html#module). Cio significa che i modelli sono compatibili con l'uso di ciascun di framework.
+Il prossimo passo e di creare [modello](main_classes/models). Il modello - vagamente riferito anche come architettura - definisce cosa ogni strato deve fare e quali operazioni stanno succedendo. Attributi come `num_hidden_layers` provenienti dalla configurazione sono usati per definire l'architettura. Ogni modello condivide la classe base [`PreTrainedModel`] e alcuni metodi comuni come il ridimensionamento degli input embeddings e la soppressione delle self-attention heads . Inoltre, tutti i modelli sono la sottoclasse di [`torch.nn.Module`](https://pytorch.org/docs/stable/generated/torch.nn.Module.html), [`tf.keras.Model`](https://www.tensorflow.org/api_docs/python/tf/keras/Model) o [`flax.linen.Module`](https://flax.readthedocs.io/en/latest/api_reference/flax.linen/module.html). Cio significa che i modelli sono compatibili con l'uso di ciascun di framework.
 
 <frameworkcontent>
 <pt>

docs/source/ja/create_a_model.md

@@ -114,7 +114,7 @@ Once you are satisfied with your model configuration, you can save it with [`Pre
 
 次のステップは、[モデル](main_classes/models)を作成することです。モデル（アーキテクチャとも緩く言われることがあります）は、各レイヤーが何をしているか、どの操作が行われているかを定義します。構成からの `num_hidden_layers` のような属性はアーキテクチャを定義するために使用されます。
 すべてのモデルは [`PreTrainedModel`] をベースクラスとし、入力埋め込みのリサイズやセルフアテンションヘッドのプルーニングなど、共通のメソッドがいくつかあります。
-さらに、すべてのモデルは [`torch.nn.Module`](https://pytorch.org/docs/stable/generated/torch.nn.Module.html)、[`tf.keras.Model`](https://www.tensorflow.org/api_docs/python/tf/keras/Model)、または [`flax.linen.Module`](https://flax.readthedocs.io/en/latest/flax.linen.html#module) のいずれかのサブクラスでもあります。つまり、モデルはそれぞれのフレームワークの使用法と互換性があります。
+さらに、すべてのモデルは [`torch.nn.Module`](https://pytorch.org/docs/stable/generated/torch.nn.Module.html)、[`tf.keras.Model`](https://www.tensorflow.org/api_docs/python/tf/keras/Model)、または [`flax.linen.Module`](https://flax.readthedocs.io/en/latest/api_reference/flax.linen/module.html) のいずれかのサブクラスでもあります。つまり、モデルはそれぞれのフレームワークの使用法と互換性があります。
 
 <frameworkcontent>
 <pt>

docs/source/ja/main_classes/processors.md

@@ -86,7 +86,7 @@ QQP、QNLI、RTE、WNLI。
 
 テストセットにはゴールドラベルが付いているため、評価はテストセットで行われますのでご了承ください。
 
-これらのプロセッサを使用する例は、[run_xnli.py](https://github.com/huggingface/transformers/tree/main/examples/legacy/text-classification/run_xnli.py) スクリプトに示されています。
+これらのプロセッサを使用する例は、[run_xnli.py](https://github.com/huggingface/transformers/tree/main/examples/pytorch/text-classification/run_xnli.py) スクリプトに示されています。
 
 ## SQuAD
 

docs/source/ja/perf_train_gpu_many.md

@@ -242,7 +242,7 @@ ZeROがモデルの重みを分割する方法に注意を払うと、これは
 
 Implementations:
 
-- [DeepSpeed](https://www.deepspeed.ai/features/#the-zero-redundancy-optimizer) ZeRO-DP stages 1+2+3
+- [DeepSpeed](https://www.deepspeed.ai/tutorials/zero/) ZeRO-DP stages 1+2+3
 - [`transformers` integration](main_classes/trainer#trainer-integrations)
 
 
@@ -360,7 +360,7 @@ by [@anton-l](https://github.com/anton-l)。
 SageMakerは、より効率的な処理のためにTPとDPを組み合わせて使用します。
 
 代替名：
-- [DeepSpeed](https://github.com/microsoft/DeepSpeed)はこれを「テンソルスライシング」と呼びます。詳細は[DeepSpeedの特徴](https://www.deepspeed.ai/features/#model-parallelism)をご覧ください。
+- [DeepSpeed](https://github.com/microsoft/DeepSpeed)はこれを「テンソルスライシング」と呼びます。詳細は[DeepSpeedの特徴](https://www.deepspeed.ai/training/#model-parallelism)をご覧ください。
 
 実装例:
 - [Megatron-LM](https://github.com/NVIDIA/Megatron-LM)には、モデル固有の内部実装があります。