DSE-QWen2-2b-MRL-V1 is a bi-encoder model designed to encode document screenshots into dense vectors for document retrieval. The Document Screenshot Embedding (DSE) approach captures documents in their original visual format, preserving all information such as text, images, and layout, thus avoiding tedious parsing and potential information loss. DSE aims to provide a generalizable embedding model for Text, PDF documents, Webpage, Slides retrieval.

For example, DSE-QWen2-2b-MRL-V1 achieves 85.8 nDCG@5 on ViDoRE leaderboard.

Note:

QWen vision encoder may take high GPU memory if the input image is large. Adjust 'resized_height':680 , 'resized_width':680 (see below) to fit VRAM based on GPU resources.

How to Use the Model

To support better effectiveness--efficiency trade-off, this checkpoint is trained to support:

Flexible representation dimension.
Flexible input image size.

Load the Model and Processor

python

import torch
from transformers import AutoProcessor, Qwen2VLForConditionalGeneration
from qwen_vl_utils import process_vision_info

min_pixels = 1*28*28
max_pixels = 2560*28*28

processor = AutoProcessor.from_pretrained("MrLight/dse-qwen2-2b-mrl-v1", min_pixels=min_pixels, max_pixels=max_pixels)
model = Qwen2VLForConditionalGeneration.from_pretrained('MrLight/dse-qwen2-2b-mrl-v1', attn_implementation="flash_attention_2", torch_dtype=torch.bfloat16).to('cuda:0').eval()
processor.tokenizer.padding_side = "left"
model.padding_side = "left"

def get_embedding(last_hidden_state: torch.Tensor, dimension: int) -> torch.Tensor:
    reps = last_hidden_state[:, -1]
    reps = torch.nn.functional.normalize(reps[:, :dimension], p=2, dim=-1)
    return reps

Encode Text Query

python

from PIL import Image
queries = ["Where can we see Llama?", "What is the LLaMA AI model?"]
query_messages = []
for query in queries:
    message = [
        {
            'role': 'user',
            'content': [
                {'type': 'image', 'image': Image.new('RGB', (28, 28)), 'resized_height':1 , 'resized_width':1}, # need a dummy image here for an easier process.
                {'type': 'text', 'text': f'Query: {query}'},
            ]
        }
    ]
    query_messages.append(message)
query_texts = [
    processor.apply_chat_template(msg, tokenize=False, add_generation_prompt=True) + "<|endoftext|>"
    for msg in query_messages
]
query_image_inputs, query_video_inputs = process_vision_info(query_messages)
query_inputs = processor(text=query_texts, images=query_image_inputs, videos=query_video_inputs, padding='longest', return_tensors='pt').to('cuda:0')
cache_position = torch.arange(0, len(query_texts))
query_inputs = model.prepare_inputs_for_generation(**query_inputs, cache_position=cache_position, use_cache=False)
with torch.no_grad():
  output = model(**query_inputs, return_dict=True, output_hidden_states=True)
query_embeddings = get_embedding(output.hidden_states[-1], 1536) # adjust dimensionality for efficiency trade-off, e.g. 512

Encode Document Screenshot

python

import requests
from io import BytesIO

# URLs of the images
url1 = "https://huggingface.co/Tevatron/dse-phi3-docmatix-v2/resolve/main/animal-llama.png"
url2 = "https://huggingface.co/Tevatron/dse-phi3-docmatix-v2/resolve/main/meta-llama.png"

# Download and open images
response1 = requests.get(url1)
response2 = requests.get(url2)

doc_image1 = Image.open(BytesIO(response1.content))
doc_image2 = Image.open(BytesIO(response2.content))

doc_images = [doc_image1, doc_image2]
doc_messages = []
for doc in doc_images:
    message = [
        {
            'role': 'user',
            'content': [
                {'type': 'image', 'image': doc}, #'resized_height':680 , 'resized_width':680} # adjust the image size for efficiency trade-off
                {'type': 'text', 'text': 'What is shown in this image?'}
            ]
        }
    ]
    doc_messages.append(message)
doc_texts = [
    processor.apply_chat_template(msg, tokenize=False, add_generation_prompt=True) + "<|endoftext|>"
    for msg in doc_messages
]
doc_image_inputs, doc_video_inputs = process_vision_info(doc_messages)
doc_inputs = processor(text=doc_texts, images=doc_image_inputs, videos=doc_video_inputs, padding='longest', return_tensors='pt').to('cuda:0')
cache_position = torch.arange(0, len(doc_texts))
doc_inputs = model.prepare_inputs_for_generation(**doc_inputs, cache_position=cache_position, use_cache=False)
with torch.no_grad():
    output = model(**doc_inputs, return_dict=True, output_hidden_states=True)
doc_embeddings = get_embedding(output.hidden_states[-1], 1536) # adjust dimensionality for efficiency trade-off e.g. 512

Compute Similarity

python

from torch.nn.functional import cosine_similarity
num_queries = query_embeddings.size(0)
num_passages = doc_embeddings.size(0)

for i in range(num_queries):
    query_embedding = query_embeddings[i].unsqueeze(0)
    similarities = cosine_similarity(query_embedding, doc_embeddings)
    print(f"Similarities for Query {i+1}: {similarities.cpu().float().numpy()}")

Encode Document Text

This DSE checkpoint is warm-up with Tevatron/msmarco-passage-aug, thus the model can also effectively encode document as text input.

python

doc_texts = [
  "The llama (/ˈlɑːmə/; Spanish pronunciation: [ˈʎama] or [ˈʝama]) (Lama glama) is a domesticated South American camelid, widely used as a meat and pack animal by Andean cultures since the pre-Columbian era.",
  "Llama (acronym for Large Language Model Meta AI, and formerly stylized as LLaMA) is a family of autoregressive large language models (LLMs) released by Meta AI starting in February 2023.[2][3] The latest version is Llama 3.1, released in July 2024.[4]"
]
doc_messages = []
for doc in doc_texts:
    message = [
        {
            'role': 'user',
            'content': [
                {'type': 'image', 'image': Image.new('RGB', (28, 28)), 'resized_height':1 , 'resized_width':1}, # need a dummy image here for an easier process.
                {'type': 'text', 'text': f'Document: {doc}'}
            ]
        }
    ]
    doc_messages.append(message)
doc_texts = [
    processor.apply_chat_template(msg, tokenize=False, add_generation_prompt=True) + "<|endoftext|>"
    for msg in doc_messages
]
doc_image_inputs, doc_video_inputs = process_vision_info(doc_messages)
doc_inputs = processor(text=doc_texts, images=doc_image_inputs, videos=doc_video_inputs, padding='longest', return_tensors='pt').to('cuda:0')
cache_position = torch.arange(0, len(doc_texts))
doc_inputs = model.prepare_inputs_for_generation(**doc_inputs, cache_position=cache_position, use_cache=False)
with torch.no_grad():
    output = model(**doc_inputs, return_dict=True, output_hidden_states=True)
doc_embeddings = get_embedding(output.hidden_states[-1], 1536) # adjust dimensionality for efficiency trade-off e.g. 512

for i in range(num_queries):
    query_embedding = query_embeddings[i].unsqueeze(0)
    similarities = cosine_similarity(query_embedding, doc_embeddings)
    print(f"Similarities for Query {i+1}: {similarities.cpu().float().numpy()}")

Citation

If you find this checkpoint is helpful, please consider citing QWen2, Docmatix, ViDoRe, and our DSE work.