Fuyu Multi-image interleaved processor

[cliangyu]

What does this PR do?

Fuyu Multi-image interleaved processor. Test example:

from transformers import FuyuProcessor, FuyuForCausalLM
from PIL import Image
import requests
import torch

# load model and processor
model_id = "adept/fuyu-8b"
processor = FuyuProcessor.from_pretrained(model_id)
model = FuyuForCausalLM.from_pretrained(model_id, device_map="cuda:0", torch_dtype=torch.bfloat16)

def convert(list_of_dicts):# Convert to a dictionary of lists
    dict_of_lists = {}
    for d in list_of_dicts:
        for key, value in d.items():
            if key not in dict_of_lists:
                dict_of_lists[key] = []
            dict_of_lists[key].append(value)
    return dict_of_lists

text_prompt1 = "|IMAGESTART| Generate a coco-style caption. |IMAGESTART| Be reminded that the caption should be longer than 2000 words but shorter than 1 million words. \n"
url1 = "https://huggingface.co/adept/fuyu-8b/resolve/main/bus.png"
image1 = Image.open(requests.get(url1, stream=True).raw)

text_prompt2 = "What doesn this chart describe?\n"
url2 = "https://huggingface.co/adept/fuyu-8b/resolve/main/chart.png"
image2 = Image.open(requests.get(url2, stream=True).raw)

test_examples = [
    # {"text": "|IMAGESTART| Generate a coco-style caption. |IMAGESTART| Be reminded that the caption should be longer than 2000 words but shorter than 1 million words. \n", "images": image1}, # should assert error
    {"text": text_prompt1, "images": [image1, image2]}, # normal
    {"text": text_prompt2, "images": [image2 for i in range(40)]}, # should add indicator
    {"text": "|IMAGESTART||IMAGESTART| Generate a coco-style caption. Be reminded that the caption should be longer than 2000 words but shorter than 1 million words. \n", "images": [image1, image2]}, # normal
    {"text": " Generate a coco-style caption. Be reminded that the caption should be longer than 2000 words but shorter than 1 million words. \n|IMAGESTART||IMAGESTART|", "images": [image1, image2]}, # normal
    # {"text": " Generate a coco-style caption. Be reminded that the caption should be longer than 2000 words but shorter than 1 million words.", "images": None}, # no image, we had error with this case
    {"text": None, "images": [image1]}, # no text
    
]
inputs_to_model = processor(**convert(test_examples), return_tensors="pt", truncation=True).to("cuda:0")

Before submitting

• This PR fixes a typo or improves the docs (you can dismiss the other checks if that's the case).
• Did you read the contributor guideline,
  Pull Request section?
• Was this discussed/approved via a Github issue or the forum? Please add a link
  to it if that's the case.
• Did you make sure to update the documentation with your changes? Here are the
  documentation guidelines, and
  here are tips on formatting docstrings.
• Did you write any new necessary tests?

Who can review?

Anyone in the community is free to review the PR once the tests have passed. Feel free to tag
members/contributors who may be interested in your PR.

@amyeroberts, @ArthurZucker and @younesbelkada

[ArthurZucker]

Hey! Feel free to ping @molbap once this is ready for review and cis are green or me if you need help with the CIs!

[cliangyu]

Hi Seungyoun, Thanks for working on this. I suggest integrating the changes with mine first. Thank you! Best regards, Liangyu Chen MMLab, School of Computer Science and Engineering, Nanyang Technological University P: 65-82811955 | E: ***@***.*** https://cliangyu.com/

…

On Sat, 2 Dec 2023 at 19:41, Seungyoun, Shin ***@***.***> wrote: Hi @cliangyu <https://github.com/cliangyu>, I've developed enhancements for multi-device support in PR #27587 <#27587>, building upon your work. Before proceeding with a new PR, I'd like to discuss integrating these changes with yours. I can submit a PR to your fork or detail the changes here for your review. Please let me know your preferred approach. — Reply to this email directly, view it on GitHub <#27587 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AKYMSERPXSTHWHLG37KE6HDYHMHU3AVCNFSM6AAAAAA7RTLUBOVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMYTQMZXGEZDOOBXGM> . You are receiving this because you authored the thread.Message ID: ***@***.***>

[ArthurZucker]

btw, let's not push the .gitignore, and it might be good to think about processing in the modeling code like what we are doing in #27662 ! 🤗 cc @molbap we can benchmark anyway before anything

[github-actions]

This issue has been automatically marked as stale because it has not had recent activity. If you think this still needs to be addressed please comment on this thread.

Please note that issues that do not follow the contributing guidelines are likely to be ignored.