Merge branch 'main' into secure-amd-ci

docs/source/ar/_toctree.yml

@@ -131,8 +131,8 @@
     title: التصدير إلى TorchScript
   - local: benchmarks
     title: المعايير
-#   - local: notebooks
-#     title: دفاتر الملاحظات مع الأمثلة
+  - local: notebooks
+    title: دفاتر الملاحظات مع الأمثلة
 #   - local: community
 #     title: موارد المجتمع
   - local: troubleshooting

docs/source/en/_toctree.yml

@@ -657,6 +657,8 @@
         title: GLPN
       - local: model_doc/hiera
         title: Hiera
+      - local: model_doc/ijepa
+        title: I-JEPA
       - local: model_doc/imagegpt
         title: ImageGPT
       - local: model_doc/levit

docs/source/en/agents.md

@@ -225,7 +225,7 @@ You have access to the following tools:
 To solve the task, you must plan forward to proceed in a series of steps, in a cycle of 'Thought:', 'Code:', and 'Observation:' sequences.
 
 At each step, in the 'Thought:' sequence, you should first explain your reasoning towards solving the task, then the tools that you want to use.
-Then in the 'Code:' sequence, you shold write the code in simple Python. The code sequence must end with '/End code' sequence.
+Then in the 'Code:' sequence, you should write the code in simple Python. The code sequence must end with '/End code' sequence.
 During each intermediate step, you can use 'print()' to save whatever important information you will then need.
 These print outputs will then be available in the 'Observation:' field, for using this information as input for the next step.
 

docs/source/en/agents_advanced.md

@@ -211,7 +211,7 @@ agent.run("How many more blocks (also denoted as layers) are in BERT base encode
 
 ## Display your agent run in a cool Gradio interface
 
-You can leverage `gradio.Chatbot`to display your agent's thoughts using `stream_to_gradio`, here is an example:
+You can leverage `gradio.Chatbot` to display your agent's thoughts using `stream_to_gradio`, here is an example:
 
 ```py
 import gradio as gr

docs/source/en/autoclass_tutorial.md

@@ -138,20 +138,23 @@ Load a processor with [`AutoProcessor.from_pretrained`]:
 
 <frameworkcontent>
 <pt>
-The `AutoModelFor` classes let you load a pretrained model for a given task (see [here](model_doc/auto) for a complete list of available tasks). For example, load a model for sequence classification with [`AutoModelForSequenceClassification.from_pretrained`]:
+The `AutoModelFor` classes let you load a pretrained model for a given task (see [here](model_doc/auto) for a complete list of available tasks). For example, load a model for sequence classification with [`AutoModelForSequenceClassification.from_pretrained`].
+
+> [!WARNING]
+> By default, the weights are loaded in full precision (torch.float32) regardless of the actual data type the weights are stored in such as torch.float16. Set `torch_dtype="auto"` to load the weights in the data type defined in a model's `config.json` file to automatically load the most memory-optimal data type.
 
 ```py
 >>> from transformers import AutoModelForSequenceClassification
 
->>> model = AutoModelForSequenceClassification.from_pretrained("distilbert/distilbert-base-uncased")
+>>> model = AutoModelForSequenceClassification.from_pretrained("distilbert/distilbert-base-uncased", torch_dtype="auto")
 ```
 
 Easily reuse the same checkpoint to load an architecture for a different task:
 
 ```py
 >>> from transformers import AutoModelForTokenClassification
 
->>> model = AutoModelForTokenClassification.from_pretrained("distilbert/distilbert-base-uncased")
+>>> model = AutoModelForTokenClassification.from_pretrained("distilbert/distilbert-base-uncased", torch_dtype="auto")
 ```
 
 <Tip warning={true}>

docs/source/en/generation_strategies.md

@@ -456,6 +456,8 @@ just like in multinomial sampling. However, in assisted decoding, reducing the t
 ['Alice and Bob, a couple of friends of mine, who are both in the same office as']
 ```
 
+We recommend to install `scikit-learn` library to enhance the candidate generation strategy and achieve additional speedup.
+
 #### Universal Assisted Decoding
 
 Universal Assisted Decoding (UAD) adds support for main and assistant models with different tokenizers.

docs/source/en/index.md

@@ -168,6 +168,7 @@ Flax), PyTorch, and/or TensorFlow.
 |                         [Hiera](model_doc/hiera)                         |       ✅        |         ❌         |      ❌      |
 |                        [Hubert](model_doc/hubert)                        |       ✅        |         ✅         |      ❌      |
 |                        [I-BERT](model_doc/ibert)                         |       ✅        |         ❌         |      ❌      |
+|                        [I-JEPA](model_doc/ijepa)                         |       ✅        |         ❌         |      ❌      |
 |                       [IDEFICS](model_doc/idefics)                       |       ✅        |         ✅         |      ❌      |
 |                      [Idefics2](model_doc/idefics2)                      |       ✅        |         ❌         |      ❌      |
 |                      [Idefics3](model_doc/idefics3)                      |       ✅        |         ❌         |      ❌      |

docs/source/en/internal/generation_utils.md

@@ -436,3 +436,9 @@ A [`Constraint`] can be used to force the generation to include specific tokens
 
 [[autodoc]] SynthIDTextWatermarkDetector
     - __call__
+
+## Compile Utils
+
+[[autodoc]] CompileConfig
+    - __call__
+

docs/source/en/kv_cache.md

@@ -180,7 +180,7 @@ Fun fact: The shortest war in history was between Britain and Zanzibar on August
 
 <Tip warning={true}>
 
-Cache offloading requires a GPU and can be slower than dynamic KV cache. Use it if you are getting CUDA out of memory errors.
+Cache offloading requires a CUDA GPU and can be slower than dynamic KV cache. Use it if you are getting CUDA out of memory errors.
 
 </Tip>
 
@@ -261,6 +261,7 @@ This will use the [`~OffloadedStaticCache`] implementation instead.
 >>> tokenizer.batch_decode(out, skip_special_tokens=True)[0]
 "Hello, my name is [Your Name], and I am a [Your Profession] with [Number of Years] of"
 ```
+Cache offloading requires a CUDA GPU.
 
 
 ### Sliding Window Cache

docs/source/en/llm_optims.md

@@ -57,13 +57,13 @@ import os
 os.environ["TOKENIZERS_PARALLELISM"] = "false"  # To prevent long warnings :)
 
 tokenizer = AutoTokenizer.from_pretrained("google/gemma-2b")
-model = AutoModelForCausalLM.from_pretrained("google/gemma-2b", device_map="auto")
+model = AutoModelForCausalLM.from_pretrained("google/gemma-2b", torch_dtype="auto", device_map="auto")
 
 model.generation_config.cache_implementation = "static"
 
 model.forward = torch.compile(model.forward, mode="reduce-overhead", fullgraph=True)
 input_text = "The theory of special relativity states "
-input_ids = tokenizer(input_text, return_tensors="pt").to("cuda")
+input_ids = tokenizer(input_text, return_tensors="pt").to(model.device.type)
 
 outputs = model.generate(**input_ids)
 print(tokenizer.batch_decode(outputs, skip_special_tokens=True))
@@ -89,11 +89,11 @@ import os
 os.environ["TOKENIZERS_PARALLELISM"] = "false"  # To prevent long warnings :)
 
 tokenizer = AutoTokenizer.from_pretrained("google/gemma-2b")
-model = AutoModelForCausalLM.from_pretrained("google/gemma-2b", device_map="auto")
+model = AutoModelForCausalLM.from_pretrained("google/gemma-2b", torch_dtype="auto", device_map="auto")
 
 model.forward = torch.compile(model.forward, mode="reduce-overhead", fullgraph=True)
 input_text = "The theory of special relativity states "
-input_ids = tokenizer(input_text, return_tensors="pt").to("cuda")
+input_ids = tokenizer(input_text, return_tensors="pt").to(model.device.type)
 prompt_length = input_ids.input_ids.shape[1]
 model.generation_config.max_new_tokens = 16
 
@@ -126,14 +126,15 @@ If you want to go further down a level, the [`StaticCache`] object can also be p
 from transformers import LlamaTokenizer, LlamaForCausalLM, StaticCache, logging
 from transformers.testing_utils import CaptureLogger
 import torch
+from accelerate.test_utils.testing import get_backend
 
 prompts = [
     "Simply put, the theory of relativity states that ",
     "My favorite all time favorite condiment is ketchup.",
 ]
 
 NUM_TOKENS_TO_GENERATE = 40
-torch_device = "cuda"
+torch_device, _, _ = get_backend() # automatically detects the underlying device type (CUDA, CPU, XPU, MPS, etc.)
 
 tokenizer = LlamaTokenizer.from_pretrained("meta-llama/Llama-2-7b-hf", pad_token="</s>", padding_side="right")
 model = LlamaForCausalLM.from_pretrained("meta-llama/Llama-2-7b-hf", device_map="sequential")
@@ -201,11 +202,11 @@ import os
 os.environ["TOKENIZERS_PARALLELISM"] = "false"  # To prevent long warnings :)
 
 tokenizer = AutoTokenizer.from_pretrained("google/gemma-2b")
-model = AutoModelForCausalLM.from_pretrained("google/gemma-2b", device_map="auto")
+model = AutoModelForCausalLM.from_pretrained("google/gemma-2b", torch_dtype="auto", device_map="auto")
 
 model.generate = torch.compile(model.generate, mode="reduce-overhead", fullgraph=True)
 input_text = "The theory of special relativity states "
-input_ids = tokenizer(input_text, return_tensors="pt").to("cuda")
+input_ids = tokenizer(input_text, return_tensors="pt").to(model.device.type)
 
 outputs = model.generate(**input_ids)
 print(tokenizer.batch_decode(outputs, skip_special_tokens=True))
@@ -241,13 +242,14 @@ Enable speculative decoding by loading an assistant model and passing it to the
 ```py
 from transformers import AutoModelForCausalLM, AutoTokenizer
 import torch
+from accelerate.test_utils.testing import get_backend
 
-device = "cuda" if torch.cuda.is_available() else "cpu"
+device, _, _ = get_backend() # automatically detects the underlying device type (CUDA, CPU, XPU, MPS, etc.)
 
 tokenizer = AutoTokenizer.from_pretrained("facebook/opt-1.3b")
 inputs = tokenizer("Einstein's theory of relativity states", return_tensors="pt").to(device)
 
-model = AutoModelForCausalLM.from_pretrained("facebook/opt-1.3b").to(device)
+model = AutoModelForCausalLM.from_pretrained("facebook/opt-1.3b", torch_dtype="auto").to(device)
 assistant_model = AutoModelForCausalLM.from_pretrained("facebook/opt-125m").to(device)
 outputs = model.generate(**inputs, assistant_model=assistant_model)
 tokenizer.batch_decode(outputs, skip_special_tokens=True)
@@ -262,13 +264,14 @@ For speculative sampling decoding, add the `do_sample` and `temperature` paramet
 ```py
 from transformers import AutoModelForCausalLM, AutoTokenizer
 import torch
+from accelerate.test_utils.testing import get_backend
 
-device = "cuda" if torch.cuda.is_available() else "cpu"
+device, _, _ = get_backend() # automatically detects the underlying device type (CUDA, CPU, XPU, MPS, etc.)
 
 tokenizer = AutoTokenizer.from_pretrained("facebook/opt-1.3b")
 inputs = tokenizer("Einstein's theory of relativity states", return_tensors="pt").to(device)
 
-model = AutoModelForCausalLM.from_pretrained("facebook/opt-1.3b").to(device)
+model = AutoModelForCausalLM.from_pretrained("facebook/opt-1.3b", torch_dtype="auto").to(device)
 assistant_model = AutoModelForCausalLM.from_pretrained("facebook/opt-125m").to(device)
 outputs = model.generate(**inputs, assistant_model=assistant_model, do_sample=True, temperature=0.7)
 print(tokenizer.batch_decode(outputs, skip_special_tokens=True))
@@ -290,13 +293,14 @@ To enable prompt lookup decoding, specify the number of tokens that should be ov
 ```py
 from transformers import AutoModelForCausalLM, AutoTokenizer
 import torch
+from accelerate.test_utils.testing import get_backend
 
-device = "cuda" if torch.cuda.is_available() else "cpu"
+device, _, _ = get_backend() # automatically detects the underlying device type (CUDA, CPU, XPU, MPS, etc.)
 
 tokenizer = AutoTokenizer.from_pretrained("facebook/opt-1.3b")
 inputs = tokenizer("The second law of thermodynamics states", return_tensors="pt").to(device)
 
-model = AutoModelForCausalLM.from_pretrained("facebook/opt-1.3b").to(device)
+model = AutoModelForCausalLM.from_pretrained("facebook/opt-1.3b", torch_dtype="auto").to(device)
 assistant_model = AutoModelForCausalLM.from_pretrained("facebook/opt-125m").to(device)
 outputs = model.generate(**inputs, prompt_lookup_num_tokens=3)
 print(tokenizer.batch_decode(outputs, skip_special_tokens=True))
@@ -311,13 +315,14 @@ For prompt lookup decoding with sampling, add the `do_sample` and `temperature`
 ```py
 from transformers import AutoModelForCausalLM, AutoTokenizer
 import torch
+from accelerate.test_utils.testing import get_backend
 
-device = "cuda" if torch.cuda.is_available() else "cpu"
+device, _, _ = get_backend() # automatically detects the underlying device type (CUDA, CPU, XPU, MPS, etc.)
 
 tokenizer = AutoTokenizer.from_pretrained("facebook/opt-1.3b")
 inputs = tokenizer("The second law of thermodynamics states", return_tensors="pt").to(device)
 
-model = AutoModelForCausalLM.from_pretrained("facebook/opt-1.3b").to(device)
+model = AutoModelForCausalLM.from_pretrained("facebook/opt-1.3b", torch_dtype="auto").to(device)
 outputs = model.generate(**inputs, prompt_lookup_num_tokens=3, do_sample=True, temperature=0.7)
 print(tokenizer.batch_decode(outputs, skip_special_tokens=True))
 ["The second law of thermodynamics states that energy cannot be created nor destroyed. It's not a"]

docs/source/en/model_doc/ijepa.md

@@ -0,0 +1,78 @@
+<!--Copyright 2024 The HuggingFace Team. All rights reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with
+the License. You may obtain a copy of the License at
+
+http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on
+an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the
+specific language governing permissions and limitations under the License.
+
+⚠️ Note that this file is in Markdown but contain specific syntax for our doc-builder (similar to MDX) that may not be
+rendered properly in your Markdown viewer.
+
+-->
+
+# I-JEPA
+
+## Overview
+
+The I-JEPA model was proposed in [Image-based Joint-Embedding Predictive Architecture](https://arxiv.org/pdf/2301.08243.pdf) by Mahmoud Assran, Quentin Duval, Ishan Misra, Piotr Bojanowski, Pascal Vincent, Michael Rabbat, Yann LeCun, Nicolas Ballas.
+I-JEPA is a self-supervised learning method that predicts the representations of one part of an image based on other parts of the same image. This approach focuses on learning semantic features without relying on pre-defined invariances from hand-crafted data transformations, which can bias specific tasks, or on filling in pixel-level details, which often leads to less meaningful representations.
+
+The abstract from the paper is the following:
+
+This paper demonstrates an approach for learning highly semantic image representations without relying on hand-crafted data-augmentations. We introduce the Image- based Joint-Embedding Predictive Architecture (I-JEPA), a non-generative approach for self-supervised learning from images. The idea behind I-JEPA is simple: from a single context block, predict the representations of various target blocks in the same image. A core design choice to guide I-JEPA towards producing semantic representations is the masking strategy; specifically, it is crucial to (a) sample tar- get blocks with sufficiently large scale (semantic), and to (b) use a sufficiently informative (spatially distributed) context block. Empirically, when combined with Vision Transform- ers, we find I-JEPA to be highly scalable. For instance, we train a ViT-Huge/14 on ImageNet using 16 A100 GPUs in under 72 hours to achieve strong downstream performance across a wide range of tasks, from linear classification to object counting and depth prediction.
+
+This model was contributed by [jmtzt](https://huggingface.co/jmtzt).
+The original code can be found [here](https://github.com/facebookresearch/ijepa).
+
+## How to use
+
+Here is how to use this model for image feature extraction:
+
+```python
+import requests
+import torch
+from PIL import Image
+from torch.nn.functional import cosine_similarity
+
+from transformers import AutoModel, AutoProcessor
+
+url_1 = "http://images.cocodataset.org/val2017/000000039769.jpg"
+url_2 = "http://images.cocodataset.org/val2017/000000219578.jpg"
+image_1 = Image.open(requests.get(url_1, stream=True).raw)
+image_2 = Image.open(requests.get(url_2, stream=True).raw)
+
+model_id = "jmtzt/ijepa_vith14_1k"
+processor = AutoProcessor.from_pretrained(model_id)
+model = AutoModel.from_pretrained(model_id)
+
+@torch.no_grad()
+def infer(image):
+    inputs = processor(image, return_tensors="pt")
+    outputs = model(**inputs)
+    return outputs.last_hidden_state.mean(dim=1)
+
+
+embed_1 = infer(image_1)
+embed_2 = infer(image_2)
+
+similarity = cosine_similarity(embed_1, embed_2)
+print(similarity)
+```
+
+## IJepaConfig
+
+[[autodoc]] IJepaConfig
+
+## IJepaModel
+
+[[autodoc]] IJepaModel
+    - forward
+
+## IJepaForImageClassification
+
+[[autodoc]] IJepaForImageClassification
+    - forward

docs/source/en/perf_infer_cpu.md

@@ -41,7 +41,7 @@ Enable BetterTransformer with the [`PreTrainedModel.to_bettertransformer`] metho
 ```py
 from transformers import AutoModelForCausalLM
 
-model = AutoModelForCausalLM.from_pretrained("bigcode/starcoder")
+model = AutoModelForCausalLM.from_pretrained("bigcode/starcoder", torch_dtype="auto")
 ```
 
 ## TorchScript