Hugging Face's logo

erax-ai
/
EraX-VL-2B-V1.5

Visual Question Answering
Transformers
Safetensors
Vietnamese
English
Chinese
qwen2_vl
image-text-to-text
erax
multimodal
erax-vl-2B
insurance
ocr
vietnamese
bcg
text-generation-inference
Inference Endpoints
Model card Files Community
5

Logo

EraX-VL-2B-V1.5

Introduction ๐ŸŽ‰

Hot on the heels of the popular EraX-VL-7B-V1.0 model, we proudly present EraX-VL-2B-V1.5. This enhanced multimodal model offers robust OCR and VQA capabilities across diverse languages ๐ŸŒ, with a significant advantage in processing Vietnamese ๐Ÿ‡ป๐Ÿ‡ณ. The EraX-VL-2B model stands out for its precise recognition capabilities across a range of documents ๐Ÿ“, including medical forms ๐Ÿฉบ, invoices ๐Ÿงพ, bills of sale ๐Ÿ’ณ, quotes ๐Ÿ“„, and medical records ๐Ÿ’Š. This functionality is expected to be highly beneficial for hospitals ๐Ÿฅ, clinics ๐Ÿ’‰, insurance companies ๐Ÿ›ก๏ธ, and other similar applications ๐Ÿ“‹. Built on the solid foundation of the Qwen/Qwen2-VL-2B-Instruct[1], which we found to be of high quality and fluent in Vietnamese, EraX-VL-2B has been fine-tuned to enhance its performance. We plan to continue improving and releasing new versions for free, along with sharing performance benchmarks in the near future.

One standing-out feature of EraX-VL-2B-V1.5 is the capability to do multi-turn Q&A with reasonable reasoning capability at its small size of only +2 billions parameters.

NOTA BENE:

  • EraX-VL-2B-V1.5 is NOT a typical OCR-only tool likes Tesseract but is a Multimodal LLM-based model. To use it effectively, you may have to twist your prompt carefully depending on your tasks.
  • This model was NOT finetuned with medical (X-ray) dataset or car accidences (yet). Stay tune for updated version coming up sometime 2025.

EraX-VL-2B-V1.5 is a young and tiny member of our EraX's Lร nhGPT collection of LLM models.

Benchmarks ๐Ÿ“Š

๐Ÿ† LeaderBoard

Models Open-Source VI-MTVQA
EraX-VL-7B-V1.5 ๐Ÿฅ‡ (soon) 47.2
Qwen2-VL 72B ๐Ÿฅˆ โœ˜ 41.6
ViGPT-VL ๐Ÿฅ‰ โœ˜ 39.1
EraX-VL-2B-V1.5 โœ… 38.2
EraX-VL-7B-V1 โœ… 37.6
Vintern-1B-V2 โœ… 37.4
Qwen2-VL 7B โœ… 30.0
Claude3 Opus โœ˜ 29.1
GPT-4o mini โœ˜ 29.1
GPT-4V โœ˜ 28.9
Gemini Ultra โœ˜ 28.6
InternVL2 76B โœ… 26.9
QwenVL Max โœ˜ 23.5
Claude3 Sonnet โœ˜ 20.8
QwenVL Plus โœ˜ 18.1
MiniCPM-V2.5 โœ… 15.3

The test code for evaluating models in the paper can be found in: EraX-JS-Company/EraX-MTVQA-Benchmark

API trial ๐ŸŽ‰

Please contact [email protected] for API access inquiry.

Examples ๐Ÿงฉ

1. OCR - Optical Character Recognition for Multi-Images

Example 01: Citizen identification card

Front View

Front View

Back View

Back View

Source: Google Support 

{
  "Sแป‘ thแบป":"037094012351"
  "Hแป vร  tรชn":"TRแปŠNH QUANG DUY"
  "Ngร y sinh":"04/09/1994"
  "Giแป›i tรญnh":"Nam"
  "Quแป‘c tแป‹ch":"Viแป‡t Nam"
  "Quรช quรกn / Place of origin":"Tรขn Thร nh, Kim Sฦกn, Ninh Bรฌnh"
  "Nฦกi thฦฐแปng trรบ / Place of residence":"Xรณm 6 Tรขn Thร nh, Kim Sฦกn, Ninh Bรฌnh"
  "Cรณ giรก trแป‹ ฤ‘แบฟn":"04/09/2034"
  "ฤแบทc ฤ‘iแปƒm nhรขn dแบกng / Personal identification":"seo chแบฅm c:1cm trรชn ฤ‘uรดi mแบฏt trรกi"
  "Cแปฅc trฦฐแปŸng cแปฅc cแบฃnh sรกt quแบฃn lรฝ hร nh chรญnh vแป trแบญt tแปฑ xรฃ hแป™i":"Nguyแป…n Quแป‘c Hรนng"
  "Ngร y cแบฅp":"10/12/2022"
}

Example 01: Identity Card

Front View

Front View

Back View

Back View

Source: Internet 

{
  "Sแป‘":"272737384"
  "Hแป tรชn":"PHแบ M NHแบฌT TRฦฏแปœNG"
  "Sinh ngร y":"08-08-2000"
  "Nguyรชn quรกn":"Tiแปn Giang"
  "Nฦกi ฤKHK thฦฐแปng trรบ":"393, Tรขn Xuรขn, Bแบฃo Bรฌnh, Cแบฉm Mแปน, ฤแป“ng Nai"
  "Dรขn tแป™c":"Kinh"
  "Tรดn giรกo":"Khรดng"
  "ฤแบทc ฤ‘iแปƒm nhแบญn dแบกng":"Nแป‘t ruแป“i c.3,5cm trรชn sau cรกnh mลฉi phแบฃi."
  "Ngร y cแบฅp":"30 thรกng 01 nฤƒm 2018"
  "Giรกm ฤ‘แป‘c CA":"T.BรŒNH ฤแปŠNH"
}

Example 02: Driver's License

Front View

Front View

Back View

Back View

Source: Bรกo Phรกp luแบญt 

{
  "No.":"400116012313"
  "Fullname":"NGUYแป„N Vฤ‚N DลจNG"
  "Date_of_birth":"08/06/1979"
  "Nationality":"VIแป†T NAM"
  "Address":"X. Quแปณnh Hแบงu, H. Quแปณnh Lฦฐu, T. Nghแป‡ An
  Nghแป‡ An, ngร y/date 23 thรกng/month 04 nฤƒm/year 2022"
  "Hang_Class":"FC"
  "Expires":"23/04/2027"
  "Place_of_issue":"Nghแป‡ An"
  "Date_of_issue":"ngร y/date 23 thรกng/month 04 nฤƒm/year 2022"
  "Signer":"Trแบงn Anh Tuแบฅn"
  "Cรกc loแบกi xe ฤ‘ฦฐแปฃc phรฉp":"ร” tรด hแบกng C kรฉo rฦกmoรณc, ฤ‘แบงu kรฉo kรฉo sฦกmi rฦกmoรณc vร  xe hแบกng B1, B2, C, FB2 (Motor vehicle of class C with a trailer, semi-trailer truck and vehicles of classes B1, B2, C, FB2)"
  "Mรฃ sแป‘":""
}

Example 03: Vehicle Registration Certificate

Source: Bรกo Vietnamnet 

{
  "Tรชn chแปง xe":"NGUYแป„N Tร”N NHUแบฌN"
  "ฤแป‹a chแป‰":"KE27 Kp3 P.TTTรขy Q7"
  "Nhรฃn hiแป‡u":"HONDA"
  "Sแป‘ loแบกi":"DYLAN"
  "Mร u sฦกn":"Trแบฏng"
  "Sแป‘ ngฦฐแปi ฤ‘ฦฐแปฃc phรฉp chแปŸ":"02"
  "Nguแป“n gแป‘c":"Xe nhแบญp mแป›i"
  "Biแปƒn sแป‘ ฤ‘ฤƒng kรฝ":"59V1-498.89"
  "ฤฤƒng kรฝ lแบงn ฤ‘แบงu ngร y":"08/06/2004"
  "Sแป‘ mรกy":"F03E-0057735"
  "Sแป‘ khung":"5A04F-070410"
  "Dung tรญch":"152"
  "Quแบฃn lรฝ":"TRฦฏแปžNG CA QUแบฌN"
  "Thฦฐแปฃng tรก":"Trแบงn Vฤƒn Hiแปƒu"
}

Example 04: Birth Certificate

Source: https://congchung247.com.vn 

{
    "name": "NGUYแป„N NAM PHฦฏฦ NG",
    "gender": "Nแปฏ",
    "date_of_birth": "08/6/2011",
    "place_of_birth": "Bแป‡nh viแป‡n Viแป‡t - Phรกp Hร  Nแป™i",
    "nationality": "Viแป‡t Nam",
    "father_name": "Nguyแป…n Ninh Hแป“ng Quang",
    "father_dob": "1980",
    "father_address": "309 nhร  E2 Bแบกch Khoa - Hai Bร  Trฦฐng - Hร  Nแป™i",
    "mother_name": "Phแบกm Thรนy Trang",
    "mother_dob": "1984",
    "mother_address": "309 nhร  E2 Bแบกch Khoa - Hai Bร  Trฦฐng - Hร  Nแป™i",
    "registration_place": "UBND phฦฐแปng Bแบกch Khoa - Quแบญn Hai Bร  Trฦฐng - Hร  Nแป™i",
    "registration_date": "05/8/2011",
    "registration_ralation": "cha",
    "notes": None,
    "certified_by": "Nguyแป…n Thแป‹ Kim Hoa"
}

Quickstart ๐ŸŽฎ

Install the necessary packages:

python -m pip install git+https://github.com/huggingface/transformers accelerate
python -m pip install qwen-vl-utils
pip install flash-attn --no-build-isolation

Then you can use EraX-VL-2B-V1.5 like this:

import os
import base64
import json

import cv2
import numpy as np
import matplotlib.pyplot as  plt

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

model_path = "erax/EraX-VL-2B-V1.5"

model = Qwen2VLForConditionalGeneration.from_pretrained(
    model_path,
    torch_dtype=torch.bfloat16,
    attn_implementation="eager", # replace with "flash_attention_2" if your GPU is Ampere architecture
    device_map="auto"
)

tokenizer =  AutoTokenizer.from_pretrained(model_path)
# processor = AutoProcessor.from_pretrained(model_path)

min_pixels = 256 * 28 * 28
max_pixels = 1280 * 28 * 28
processor = AutoProcessor.from_pretrained(
     model_path,
     min_pixels=min_pixels,
     max_pixels=max_pixels,
 )

image_path ="image.jpg"

with open(image_path, "rb") as f:
    encoded_image = base64.b64encode(f.read())
decoded_image_text = encoded_image.decode('utf-8')
base64_data = f"data:image;base64,{decoded_image_text}"

messages = [
    {
        "role": "user",
        "content": [
            {
                "type": "image",
                "image": base64_data,
            },
            {
                "type": "text",
                "text": "Trรญch xuแบฅt thรดng tin nแป™i dung tแปซ hรฌnh แบฃnh ฤ‘ฦฐแปฃc cung cแบฅp."
            },
        ],
    }
]

# Prepare prompt
tokenized_text = processor.apply_chat_template(
    messages, tokenize=False, add_generation_prompt=True
)

image_inputs, video_inputs = process_vision_info(messages)

inputs = processor(
    text=[ tokenized_text],
    images=image_inputs,
    videos=video_inputs,
    padding=True,
    return_tensors="pt",
)
inputs = inputs.to("cuda")

# Generation configs
generation_config =  model.generation_config
generation_config.do_sample   = True
generation_config.temperature = 1.0
generation_config.top_k       = 1
generation_config.top_p       = 0.9
generation_config.min_p       = 0.1
generation_config.best_of     = 5
generation_config.max_new_tokens     = 2048
generation_config.repetition_penalty = 1.06

# Inference
generated_ids = model.generate(**inputs, generation_config=generation_config)
generated_ids_trimmed = [
    out_ids[len(in_ids) :] for in_ids, out_ids in zip(inputs.input_ids, generated_ids)
]
output_text = processor.batch_decode(
    generated_ids_trimmed, skip_special_tokens=True, clean_up_tokenization_spaces=False
)

print(output_text[0])

References ๐Ÿ“‘

[1] Qwen team. Qwen2-VL. 2024.

[2] Bai, Jinze, et al. "Qwen-VL: A Versatile Vision-Language Model for Understanding, Localization, Text Reading, and Beyond." arXiv preprint arXiv:2308.12966 (2023).

[4] Yang, An, et al. "Qwen2 technical report." arXiv preprint arXiv:2407.10671 (2024).

[5] Chen, Zhe, et al. "Internvl: Scaling up vision foundation models and aligning for generic visual-linguistic tasks." Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2024.

[6] Chen, Zhe, et al. "How far are we to gpt-4v? closing the gap to commercial multimodal models with open-source suites." arXiv preprint arXiv:2404.16821 (2024).

[7] Tran, Chi, and Huong Le Thanh. "LaVy: Vietnamese Multimodal Large Language Model." arXiv preprint arXiv:2404.07922 (2024).

Contact ๐Ÿค

  • For correspondence regarding this work or inquiry for API trial, please contact Nguyแป…n Anh Nguyรชn at [email protected].
  • Follow us on EraX Github
Downloads last month
412
Safetensors
Model size
2.21B params
Tensor type
F32
ยท
Inference Examples
Visual Question Answering
This model does not have enough activity to be deployed to Inference API (serverless) yet. Increase its social visibility and check back later, or deploy to Inference Endpoints (dedicated) instead.

Model tree for erax-ai/EraX-VL-2B-V1.5

Base model

Qwen/Qwen2-VL-2B
Finetuned
Qwen/Qwen2-VL-2B-Instruct
Finetuned
(57)
this model
Quantizations
2 models

Collection including erax-ai/EraX-VL-2B-V1.5