VictorSanh's picture
there you go
5b86ca4
|
raw
history blame
6.58 kB
metadata
license: apache-2.0
language:
  - en
datasets:
  - HuggingFaceM4/WebSight
tags:
  - code

Model Description

This model converts screenshots of website components into HTML/Tailwind CSS codes.

It is based on an early checkpoint of our forthcoming vision-language foundation model, which has been further fine-tuned with DoRA using the Websight-v1 dataset.

The base model is built upon Mistral-7B and SigLIP-SO400M, and uses the Patch n’ Pack strategy to preserve the original aspect ratio of the input images, with a resolution of up to 980 pixels for each side. Further insights into the model’s architecture and its training process will be detailed upon its release.

The goal of open-sourcing the WebSight dataset along with the model Sightseer is to kick off an effort to develop improved models capable of converting a website screenshot into actual code.

Try out the demo!

Code snippet

The code snippet demonstrates how to perform batched generation to convert screenshots of websites into corresponding HTML + Tailwind code.

Note that the logic to process, and pad inputs will be encapsulated into a user-friendly processor upon the release of our vision and language model.

import torch
import requests

from datasets import load_dataset
from io import BytesIO
from transformers import AutoModelForCausalLM, AutoProcessor
from PIL import Image

from transformers.image_utils import to_numpy_array, PILImageResampling, ChannelDimension
from transformers.image_transforms import resize, to_channel_dimension_format


DEVICE = torch.device("cuda")
PROCESSOR = AutoProcessor.from_pretrained(
    "HuggingFaceM4/Sightseer",
    token=API_TOKEN,
)
MODEL = AutoModelForCausalLM.from_pretrained(
    "HuggingFaceM4/Sightseer",
    token=API_TOKEN,
    trust_remote_code=True,
    torch_dtype=torch.bfloat16,
).to(DEVICE)
image_seq_len = MODEL.config.perceiver_config.resampler_n_latents
BOS_TOKEN = PROCESSOR.tokenizer.bos_token
BAD_WORDS_IDS = PROCESSOR.tokenizer(["<image>", "<fake_token_around_image>"], add_special_tokens=False).input_ids


def convert_to_rgb(image):
    if image.mode == "RGB":
        return image

    image_rgba = image.convert("RGBA")
    background = Image.new("RGBA", image_rgba.size, (255, 255, 255))
    alpha_composite = Image.alpha_composite(background, image_rgba)
    alpha_composite = alpha_composite.convert("RGB")
    return alpha_composite


# The processor is the same as the Idefics processor except for the BILINEAR interpolation,
# so this is a hack in order to redefine ONLY the transform method
def custom_transform(x):
    x = convert_to_rgb(x)
    x = to_numpy_array(x)

    height, width = x.shape[:2]
    aspect_ratio = width / height
    if width >= height and width > 980:
        width = 980
        height = int(width / aspect_ratio)
    elif height > width and height > 980:
        height = 980
        width = int(height * aspect_ratio)
    width = max(width, 378)
    height = max(height, 378)

    x = resize(x, (height, width), resample=PILImageResampling.BILINEAR)
    x = PROCESSOR.image_processor.rescale(x, scale=1 / 255)
    x = PROCESSOR.image_processor.normalize(
        x,
        mean=PROCESSOR.image_processor.image_mean,
        std=PROCESSOR.image_processor.image_std
    )
    x = to_channel_dimension_format(x, ChannelDimension.FIRST)
    x = torch.tensor(x)
    return x


# Create text token inputs
image_seq = '<image>' * image_seq_len
inputs = PROCESSOR.tokenizer(
    [
        f"{BOS_TOKEN}<fake_token_around_image>{image_seq}<fake_token_around_image>In this image, we see",
        f"{BOS_TOKEN}bla bla<fake_token_around_image>{image_seq}<fake_token_around_image>{image_seq}<fake_token_around_image>",
    ],
    return_tensors="pt",
    add_special_tokens=False,
    padding=True,
)


# Create pixel inputs
# We load images from WebSight, but any screenshot in the form of a PIL image will work
dataset = load_dataset("HuggingFaceM4/WebSight", streaming=True)
dataset = iter(dataset)
image1 = next(dataset)
image2 = next(dataset)
raw_images = [
    [image1],
    [image2],
]
output_images = [
    [PROCESSOR.image_processor(img, transform=custom_transform) for img in img_list]
    for img_list in raw_images
]
total_batch_size = len(output_images)
max_num_images = max([len(img_l) for img_l in output_images])
max_height = max([i.size(2) for img_l in output_images for i in img_l])
max_width = max([i.size(3) for img_l in output_images for i in img_l])
padded_image_tensor = torch.zeros(total_batch_size, max_num_images, 3, max_height, max_width)
padded_pixel_attention_masks = torch.zeros(
    total_batch_size, max_num_images, max_height, max_width, dtype=torch.bool
)
for batch_idx, img_l in enumerate(output_images):
    for img_idx, img in enumerate(img_l):
        im_height, im_width = img.size()[2:]
        padded_image_tensor[batch_idx, img_idx, :, :im_height, :im_width] = img
        padded_pixel_attention_masks[batch_idx, img_idx, :im_height, :im_width] = True

inputs["pixel_values"] = padded_image_tensor
inputs["pixel_attention_mask"] = padded_pixel_attention_masks
inputs = {k: v.to(DEVICE) for k, v in inputs.items()}

generated_ids = MODEL.generate(**inputs, bad_words_ids=BAD_WORDS_IDS, max_new_tokens=10)
generated_texts = PROCESSOR.batch_decode(generated_ids, skip_special_tokens=True)

print(generated_texts)

Model Details

License

The model is built on top of two pre-trained models: SigLIP and mistralai/Mistral-7B-v0.1, which are delivered under an Apache-2.0 license. As such, users should comply with the licenses of these models.

The two pre-trained models are connected to each other with newly initialized parameters that we train. These are not based on any of the two base frozen models forming the composite model. We release the additional weights we trained under an Apache-2.0 license.