SUMIN/HuggingFace_transformer
1
0
Fork 0
Code Issues Pull Requests Actions 7 Packages Projects Releases Wiki Activity

Add Idefics2 (#30253)

Browse Source 
* Initial add model additions

* Test

* All weights loading

* Can perform full forward pass

* Local and remote the same

* Matching local and remote

* Fixup

* Idefics2Model importable; fixup docstrings

* Don't skip by default

* Remove deprecated use_resampler arg

* Remove self.config

* DecoupledLinear takes config

* Tidy up

* Enable eager attention and tidy up

* Most tests passing

* Update for batch of processed images

* Add image processor

* Update doc pages

* Update conversion script

* Remove erroneous breakpoint

* Remove accidendtal spelling change

* Update to reflect changes on hub - make generate work

* Fix up

* Image processor tests

* Update tests

* Add a processor

* Add a processor

* Update convert script

* Update modeling file - remove fixmes

* Bug fix

* Add processing test

* Use processor

* Fix up

* Update src/transformers/models/idefics2/modeling_idefics2.py

Co-authored-by: Victor SANH <victorsanh@gmail.com>

* Update src/transformers/models/idefics2/modeling_idefics2.py

Co-authored-by: Victor SANH <victorsanh@gmail.com>

* Fix test

* Update config - PR comments and defaults align with checkpoint

* Reviewer comments

* Add copied froms for flahs attention

* Update src/transformers/models/idefics2/modeling_idefics2.py

Co-authored-by: Victor SANH <victorsanh@gmail.com>

* Apply suggestions from code review

Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com>

* Remove qk_layer_norm and freeze_layers functionality

* Fix

* Remove freeze_layer options from config

* Sync with upstream main

* Fix attention shapes siglip

* Remove Llava-next refs - TO REBASE

* Use AutoModel for text model

* Add comment to explain vision embeddings

* Fix issue with tie_word_embeddings

* Address review comments

* Fix and fix up

* Chat templates for idefics

* Fix copies

* Fix

* Add layer norms to FA2

* Fix tests

* Apply suggestions from code review

Co-authored-by: Victor SANH <victorsanh@gmail.com>

* Fix

* Review comments

* Update src/transformers/models/idefics2/modeling_idefics2.py

Co-authored-by: Victor SANH <victorsanh@gmail.com>

* Update inputs merger

* Merge weights in correct order

* Update convert script

* Update src/transformers/models/idefics2/processing_idefics2.py

Co-authored-by: Victor SANH <victorsanh@gmail.com>

* Update template

* Model code examples (fix idefics too)

* More review comments

* Tidy up

* Update processing

* Fix attention mask preparation

* Update inputs_merger inputs

* Vectorize inputs_merger

* Update src/transformers/models/idefics2/__init__.py

Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com>

* Update src/transformers/models/idefics2/modeling_idefics2.py

* Review comments

* saying bye to the `qk_layer_norms`

* Simplify

* Update latents

* Remove erroneuous readme changes

* Return images when applying chat template

* Fix bug - prompt images are for a single sample

* Update src/transformers/models/idefics2/modeling_idefics2.py

* image splitting

* fix test

* some more comment

* some comment

* Apply suggestions from code review

Co-authored-by: amyeroberts <22614925+amyeroberts@users.noreply.github.com>

* Update src/transformers/models/idefics2/image_processing_idefics2.py

Co-authored-by: amyeroberts <22614925+amyeroberts@users.noreply.github.com>

* Update processor

* Update model tests

* Update src/transformers/models/idefics2/processing_idefics2.py

Co-authored-by: Victor SANH <victorsanh@gmail.com>

* Update src/transformers/models/idefics2/processing_idefics2.py

Co-authored-by: Victor SANH <victorsanh@gmail.com>

* Don't add BOS in template

* Update src/transformers/models/idefics2/processing_idefics2.py

Co-authored-by: Victor SANH <victorsanh@gmail.com>

* Remove index in examples

* Update tests to reflect #13

* Update src/transformers/models/idefics2/processing_idefics2.py

Co-authored-by: Victor SANH <victorsanh@gmail.com>

* PR comment - consistent typing

* Update readme and model doc

* Update docs

* Update checkpoint references

* Update examples

* Fix and update tests

* Small addition

* Update tests - remove copied from as no ignore placement copy could be found

* Update example

* small fixes

* Update docs/source/en/model_doc/idefics2.md

Co-authored-by: Victor SANH <victorsanh@gmail.com>

* Update docs/source/en/model_doc/idefics2.md

Co-authored-by: Victor SANH <victorsanh@gmail.com>

* Update README.md

Co-authored-by: Victor SANH <victorsanh@gmail.com>

* Connector model as bridge

* Fix up

* Fix up

* Don't pass model inputs for generation kwargs update

* IDEFICS-2 -> Idefics2

* Remove config archive name

* IDEFICS-2 -> Idefics2

* Add back llava-next

* Update readmes

* Add requirements for processor tester

* Use custom convert_to_rgb to avoid possible BC

* Fix doc example

* Fix doc example

* Skip model doc tests - as model to large

* More doc example - account for image splitting

* Update src/transformers/image_transforms.py

* Fix config doctest

---------

Co-authored-by: Pablo Montalvo <39954772+molbap@users.noreply.github.com>
Co-authored-by: ArthurZucker <arthur.zucker@gmail.com>
Co-authored-by: Victor SANH <victorsanh@gmail.com>
Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com>
This commit is contained in:
amyeroberts
2024-04-15 17:03:03 +01:00
committed by GitHub
parent 667939a2d3
commit 6b78360e6d
41 changed files with 4692 additions and 38 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
docs/source/en/_toctree.yml
View File

@@ -738,6 +738,8 @@
title: GroupViT
- local: model_doc/idefics
title: IDEFICS
- local: model_doc/idefics2
title: Idefics2
- local: model_doc/instructblip
title: InstructBLIP
- local: model_doc/kosmos-2

1
docs/source/en/index.md
View File

@@ -160,6 +160,7 @@ Flax), PyTorch, and/or TensorFlow.
| [Hubert](model_doc/hubert) | ✅ | ✅ | ❌ |
| [I-BERT](model_doc/ibert) | ✅ | ❌ | ❌ |
| [IDEFICS](model_doc/idefics) | ✅ | ❌ | ❌ |
| [Idefics2](model_doc/idefics2) | ✅ | ❌ | ❌ |
| [ImageGPT](model_doc/imagegpt) | ✅ | ❌ | ❌ |
| [Informer](model_doc/informer) | ✅ | ❌ | ❌ |
| [InstructBLIP](model_doc/instructblip) | ✅ | ❌ | ❌ |

98
docs/source/en/model_doc/idefics2.md Normal file
View File

@@ -0,0 +1,98 @@
<!--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.
-->
# Idefics2
## Overview
The Idefics2 model was created by the [Hugging Face M4](https://huggingface.co/HuggingFaceM4) team and authored by Léo Tronchon, Hugo Laurencon, Victor Sanh.
The accompanying blog post can be found [here](https://huggingface.co/blog/idefics2).
Idefics2 is an open multimodal model that accepts arbitrary sequences of image and text inputs and produces text
outputs. The model can answer questions about images, describe visual content, create stories grounded on multiple
images, or simply behave as a pure language model without visual inputs. It improves upon IDEFICS-1, notably on
document understanding, OCR, or visual reasoning. Idefics2 is lightweight (8 billion parameters) and treats
images in their native aspect ratio and resolution, which allows for varying inference efficiency.
Tips:
- Each sample can contain multiple images, and the number of images can vary between samples. The processor will pad the inputs to the maximum number of images in a batch for input to the model.
- The processor has a `do_image_splitting` option. If `True`, each input image will be split into 4 sub-images, and concatenated with the original to form 5 images. This is useful for increasing model performance. Make sure `processor.image_processor.do_image_splitting` is set to `False` if the model was not trained with this option.
- `text` passed to the processor should have the `<image>` tokens where the images should be inserted. And `<end_of_utterance>` at the end of each utterance if the text is a chat message.
- The processor has its own `apply_chat_template` method to convert chat messages to text that can then be passed as `text` to the processor.
Example of how to use the processor on chat messages:
```python
import requests
from PIL import Image
from transformers import Idefics2Processor, Idefics2ForConditionalGeneration
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)
images = [image_1, image_2]
messages = [{
"role": "user",
"content": [
{"type": "text", "text": "What’s the difference between these two images?"},
{"type": "image"},
{"type": "image"},
],
}]
processor = Idefics2Processor.from_pretrained("HuggingFaceM4/idefics2-8b")
model = Idefics2ForConditionalGeneration.from_pretrained("HuggingFaceM4/idefics2-8b")
text = processor.apply_chat_template(messages)
# "User: What’s the difference between these two images?<image><image><end_of_utterance>\n"
print(text)
inputs = processor(images=images, text=text)
generated_text = model.generate(**inputs)
```
This model was contributed by [amyeroberts](https://huggingface.co/amyeroberts).
The original code can be found [here](https://huggingface.co/HuggingFaceM4/idefics2).
## Idefics2Config
[[autodoc]] Idefics2Config
## Idefics2Model
[[autodoc]] Idefics2Model
- forward
## Idefics2ForConditionalGeneration
[[autodoc]] Idefics2ForConditionalGeneration
- forward
## Idefics2ImageProcessor
[[autodoc]] Idefics2ImageProcessor
- preprocess
## Idefics2Processor
[[autodoc]] Idefics2Processor
- __call__

11
docs/source/en/perf_infer_gpu_one.md
View File

@@ -47,6 +47,7 @@ FlashAttention-2 is currently supported for the following architectures:
* [GPTNeo](https://huggingface.co/docs/transformers/model_doc/gpt_neo#transformers.GPTNeoModel)
* [GPTNeoX](https://huggingface.co/docs/transformers/model_doc/gpt_neox#transformers.GPTNeoXModel)
* [GPT-J](https://huggingface.co/docs/transformers/model_doc/gptj#transformers.GPTJModel)
* [Idefics2](https://huggingface.co/docs/transformers/model_doc/idefics2#transformers.Idefics2Model)
* [Falcon](https://huggingface.co/docs/transformers/model_doc/falcon#transformers.FalconModel)
* [Llama](https://huggingface.co/docs/transformers/model_doc/llama#transformers.LlamaModel)
* [Llava](https://huggingface.co/docs/transformers/model_doc/llava)
@@ -96,8 +97,8 @@ model_id = "tiiuae/falcon-7b"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
model_id,
torch_dtype=torch.bfloat16,
model_id,
torch_dtype=torch.bfloat16,
attn_implementation="flash_attention_2",
)
```
@@ -109,7 +110,7 @@ FlashAttention-2 can only be used when the model's dtype is `fp16` or `bf16`. Ma
<br>
You can also set `use_flash_attention_2=True` to enable FlashAttention-2 but it is deprecated in favor of `attn_implementation="flash_attention_2"`.
</Tip>
FlashAttention-2 can be combined with other optimization techniques like quantization to further speedup inference. For example, you can combine FlashAttention-2 with 8-bit or 4-bit quantization:
@@ -123,14 +124,14 @@ tokenizer = AutoTokenizer.from_pretrained(model_id)
# load in 8bit
model = AutoModelForCausalLM.from_pretrained(
model_id,
model_id,
load_in_8bit=True,
attn_implementation="flash_attention_2",
)
# load in 4bit
model = AutoModelForCausalLM.from_pretrained(
model_id,
model_id,
load_in_4bit=True,
attn_implementation="flash_attention_2",
)