README.md

---
library_name: transformers
license: mit
language:
- vi
- en
- zh
base_model:
- OpenGVLab/InternVL2_5-1B
pipeline_tag: image-text-to-text
---

<div align="center">
  <img src="https://cdn-uploads.huggingface.co/production/uploads/6336b5c831efcb5647f00170/-G297bBqMzYvTbD6_Bkd9.png" width="500"/>
</div>

# Vintern-1B-v3.5 ❄️ (Viet-InternVL2-1B-v3.5) - The Ultimate Multimodal Solution 🌏
We introduce **Vintern-1B-v3.5**, the latest version in the Vintern series, offering significant improvements over v2 across all evaluation benchmarks. This model has been fine-tuned from **InternVL-1B-2.5**, which already good in Vietnamese tasks because it used [Viet-ShareGPT-4o-Text-VQA](https://huggingface.co/datasets/5CD-AI/Viet-ShareGPT-4o-Text-VQA) data during its fine-tuning process by the InternVL 2.5 [1] team.

<div align="center">
  <img src="https://cdn-uploads.huggingface.co/production/uploads/6336b5c831efcb5647f00170/a1V1DA1o4Gf_MJblWTz-L.png" width="500"/>
</div>
<div align="center">
  <img src="https://cdn-uploads.huggingface.co/production/uploads/6336b5c831efcb5647f00170/36jb5bgyYCoVKx3NE8Iuv.png" width="500"/>
</div>


To further enhance its performance in Vietnamese while maintaining robust capabilities on existing English datasets, **Vintern-1B-v3.5** has been fine-tuned using a vast amount of Vietnamese-specific data. This results in a model that is exceptionally powerful in text recognition, OCR, and understanding Vietnam-specific documents.

Key Features 🌟

- Top Quality for Vietnamese Texts
Vintern-1B-v3.5 is one of the best models in its class (1B parameters) for understanding and processing Vietnamese documents.

- Better Extraction and Understanding
The model is great at handling invoices, legal texts, handwriting, and tables.

- Runs on Affordable Hardware
You can run the model on Google Colab with a T4 GPU, making it easy to use without expensive devices.

- Easy to Fine-tune
The model can be customized for specific tasks with minimal effort.


## Benchmarks 📈

<div align="center">
  <img src="https://cdn-uploads.huggingface.co/production/uploads/6336b5c831efcb5647f00170/DrUCZuXuMz47uVU4zqnJ4.png" width="500"/>
</div>

| Benchmark     | InternVL2_5 1B | Vintern-1B-v2 | Vintern-1B-v3.5 |
|:-------------:|:--------------:|:-------------:|:---------------:|
| vi-MTVQA      |      24.8      |     37.4      |     41.9        |
| DocVQAtest    |      84.8      |    72.5      |      78.8       |
| MMMUval       |      40.9      |     31.3      |      32.4       |
| InfoVQAtest   |      56.0      |    38.9      |      46.4      |
| TextVQAval    |      72.0      |    64.0      |      68.2       |
| ChartQAtest   |      75.9      |    34.1      |      60.0       |
| OCRBench      |      785       |     628       |      706        |
| MMEsum        |    1950       |     1185      |      1346       |

## Examples

<div align="center">
  <img src="https://cdn-uploads.huggingface.co/production/uploads/6336b5c831efcb5647f00170/1yos0APs6laTCAGhUbN9n.png" width="300"/>
</div>

<div align="center">
  <img src="https://cdn-uploads.huggingface.co/production/uploads/6336b5c831efcb5647f00170/L5n35_3sz_Wp9fo0C7snq.png" width="300"/>
</div>



<div align="center">
  <img src="https://cdn-uploads.huggingface.co/production/uploads/6336b5c831efcb5647f00170/E6aqBwFqK38XE1LL9lF2W.png" width="500"/>
</div>

<div align="center">
  <img src="https://cdn-uploads.huggingface.co/production/uploads/6336b5c831efcb5647f00170/Lkt8YLYlDP_VByFjFQX_t.png" width="500"/>
</div>


<div align="center">
  <img src="https://cdn-uploads.huggingface.co/production/uploads/6336b5c831efcb5647f00170/wUM70ifQSpdbO_dLH1TLO.png" width="500"/>
</div>

<div align="center">
  <img src="https://cdn-uploads.huggingface.co/production/uploads/6336b5c831efcb5647f00170/yxWlGKMP7458UbtIzosUK.png" width="300"/>
</div>

<div align="center">
  <img src="https://cdn-uploads.huggingface.co/production/uploads/6336b5c831efcb5647f00170/SH7-fvyZok9Kqm1XoD4E0.png" width="500"/>
</div>

<div align="center">
  <img src="https://cdn-uploads.huggingface.co/production/uploads/6336b5c831efcb5647f00170/6gyL4ymSWyuHwfy9dVVju.png" width="500"/>
</div>


## Quickstart

Here provides a code snippet to show you how to load the tokenizer and model and how to generate contents.
To run inference using the model, follow the steps outlined in our Colab inference notebook
[![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/drive/1ZD1oB56PF0lF66RCuTVJYLTEV0tM3CFf?usp=sharing)

```python
import numpy as np
import torch
import torchvision.transforms as T
# from decord import VideoReader, cpu
from PIL import Image
from torchvision.transforms.functional import InterpolationMode
from transformers import AutoModel, AutoTokenizer

IMAGENET_MEAN = (0.485, 0.456, 0.406)
IMAGENET_STD = (0.229, 0.224, 0.225)

def build_transform(input_size):
    MEAN, STD = IMAGENET_MEAN, IMAGENET_STD
    transform = T.Compose([
        T.Lambda(lambda img: img.convert('RGB') if img.mode != 'RGB' else img),
        T.Resize((input_size, input_size), interpolation=InterpolationMode.BICUBIC),
        T.ToTensor(),
        T.Normalize(mean=MEAN, std=STD)
    ])
    return transform

def find_closest_aspect_ratio(aspect_ratio, target_ratios, width, height, image_size):
    best_ratio_diff = float('inf')
    best_ratio = (1, 1)
    area = width * height
    for ratio in target_ratios:
        target_aspect_ratio = ratio[0] / ratio[1]
        ratio_diff = abs(aspect_ratio - target_aspect_ratio)
        if ratio_diff < best_ratio_diff:
            best_ratio_diff = ratio_diff
            best_ratio = ratio
        elif ratio_diff == best_ratio_diff:
            if area > 0.5 * image_size * image_size * ratio[0] * ratio[1]:
                best_ratio = ratio
    return best_ratio

def dynamic_preprocess(image, min_num=1, max_num=12, image_size=448, use_thumbnail=False):
    orig_width, orig_height = image.size
    aspect_ratio = orig_width / orig_height

    # calculate the existing image aspect ratio
    target_ratios = set(
        (i, j) for n in range(min_num, max_num + 1) for i in range(1, n + 1) for j in range(1, n + 1) if
        i * j <= max_num and i * j >= min_num)
    target_ratios = sorted(target_ratios, key=lambda x: x[0] * x[1])

    # find the closest aspect ratio to the target
    target_aspect_ratio = find_closest_aspect_ratio(
        aspect_ratio, target_ratios, orig_width, orig_height, image_size)

    # calculate the target width and height
    target_width = image_size * target_aspect_ratio[0]
    target_height = image_size * target_aspect_ratio[1]
    blocks = target_aspect_ratio[0] * target_aspect_ratio[1]

    # resize the image
    resized_img = image.resize((target_width, target_height))
    processed_images = []
    for i in range(blocks):
        box = (
            (i % (target_width // image_size)) * image_size,
            (i // (target_width // image_size)) * image_size,
            ((i % (target_width // image_size)) + 1) * image_size,
            ((i // (target_width // image_size)) + 1) * image_size
        )
        # split the image
        split_img = resized_img.crop(box)
        processed_images.append(split_img)
    assert len(processed_images) == blocks
    if use_thumbnail and len(processed_images) != 1:
        thumbnail_img = image.resize((image_size, image_size))
        processed_images.append(thumbnail_img)
    return processed_images

def load_image(image_file, input_size=448, max_num=12):
    image = Image.open(image_file).convert('RGB')
    transform = build_transform(input_size=input_size)
    images = dynamic_preprocess(image, image_size=input_size, use_thumbnail=True, max_num=max_num)
    pixel_values = [transform(image) for image in images]
    pixel_values = torch.stack(pixel_values)
    return pixel_values

model = AutoModel.from_pretrained(
    "5CD-AI/Vintern-1B-v2",
    torch_dtype=torch.bfloat16,
    low_cpu_mem_usage=True,
    trust_remote_code=True,
).eval().cuda()
tokenizer = AutoTokenizer.from_pretrained("5CD-AI/Vintern-1B-v2", trust_remote_code=True, use_fast=False)

test_image = 'test-image.jpg'

pixel_values = load_image(test_image, max_num=12).to(torch.bfloat16).cuda()
generation_config = dict(max_new_tokens= 1024, do_sample=False, num_beams = 3, repetition_penalty=2.5)

question = '<image>\nMô tả hình ảnh một cách chi tiết.'

response, history = model.chat(tokenizer, pixel_values, question, generation_config, history=None, return_history=True)
print(f'User: {question}\nAssistant: {response}')

#question = "Câu hỏi khác ......"
#response, history = model.chat(tokenizer, pixel_values, question, generation_config, history=history, return_history=True)
#print(f'User: {question}\nAssistant: {response}')
```

## Finetune on your Data

[![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/drive/1bK6fpWfResjv9UxWoKHDStXQ8bop3a6Z?usp=sharing)


## Citation 

```
@misc{doan2024vintern1befficientmultimodallarge,
      title={Vintern-1B: An Efficient Multimodal Large Language Model for Vietnamese}, 
      author={Khang T. Doan and Bao G. Huynh and Dung T. Hoang and Thuc D. Pham and Nhat H. Pham and Quan T. M. Nguyen and Bang Q. Vo and Suong N. Hoang},
      year={2024},
      eprint={2408.12480},
      archivePrefix={arXiv},
      primaryClass={cs.LG},
      url={https://arxiv.org/abs/2408.12480}, 
}
```

## Reference

[1] Z. Chen et al., ‘Expanding performance boundaries of open-source multimodal models with model, data, and test-time scaling’, arXiv preprint arXiv:2412. 05271, 2024.