Qwen2-VL-OCR-2B-Instruct [ VL / OCR ]

The Qwen2-VL-OCR-2B-Instruct model is a fine-tuned version of Qwen/Qwen2-VL-2B-Instruct, tailored for tasks that involve Optical Character Recognition (OCR), image-to-text conversion, and math problem solving with LaTeX formatting. This model integrates a conversational approach with visual and textual understanding to handle multi-modal tasks effectively.

Key Enhancements:

• SoTA understanding of images of various resolution & ratio: Qwen2-VL achieves state-of-the-art performance on visual understanding benchmarks, including MathVista, DocVQA, RealWorldQA, MTVQA, etc.

• Understanding videos of 20min+: Qwen2-VL can understand videos over 20 minutes for high-quality video-based question answering, dialog, content creation, etc.

• Agent that can operate your mobiles, robots, etc.: with the abilities of complex reasoning and decision making, Qwen2-VL can be integrated with devices like mobile phones, robots, etc., for automatic operation based on visual environment and text instructions.

• Multilingual Support: to serve global users, besides English and Chinese, Qwen2-VL now supports the understanding of texts in different languages inside images, including most European languages, Japanese, Korean, Arabic, Vietnamese, etc.

File Name	Size	Description	Upload Status
.gitattributes	1.52 kB	Configures LFS tracking for specific model files.	Initial commit
README.md	203 Bytes	Minimal details about the uploaded model.	Updated
added_tokens.json	408 Bytes	Additional tokens used by the model tokenizer.	Uploaded
chat_template.json	1.05 kB	Template for chat-based model input/output.	Uploaded
config.json	1.24 kB	Model configuration metadata.	Uploaded
generation_config.json	252 Bytes	Configuration for text generation settings.	Uploaded
merges.txt	1.82 MB	BPE merge rules for tokenization.	Uploaded
model.safetensors	4.42 GB	Serialized model weights in a secure format.	Uploaded (LFS)
preprocessor_config.json	596 Bytes	Preprocessing configuration for input data.	Uploaded
vocab.json	2.78 MB	Vocabulary file for tokenization.	Uploaded

---

How to Use

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

# default: Load the model on the available device(s)
model = Qwen2VLForConditionalGeneration.from_pretrained(
    "prithivMLmods/Qwen2-VL-OCR-2B-Instruct", torch_dtype="auto", device_map="auto"
)

# We recommend enabling flash_attention_2 for better acceleration and memory saving, especially in multi-image and video scenarios.
# model = Qwen2VLForConditionalGeneration.from_pretrained(
#     "prithivMLmods/Qwen2-VL-OCR-2B-Instruct",
#     torch_dtype=torch.bfloat16,
#     attn_implementation="flash_attention_2",
#     device_map="auto",
# )

# default processer
processor = AutoProcessor.from_pretrained("prithivMLmods/Qwen2-VL-OCR-2B-Instruct")

# The default range for the number of visual tokens per image in the model is 4-16384. You can set min_pixels and max_pixels according to your needs, such as a token count range of 256-1280, to balance speed and memory usage.
# min_pixels = 256*28*28
# max_pixels = 1280*28*28
# processor = AutoProcessor.from_pretrained("Qwen/Qwen2-VL-2B-Instruct", min_pixels=min_pixels, max_pixels=max_pixels)

messages = [
    {
        "role": "user",
        "content": [
            {
                "type": "image",
                "image": "https://qianwen-res.oss-cn-beijing.aliyuncs.com/Qwen-VL/assets/demo.jpeg",
            },
            {"type": "text", "text": "Describe this image."},
        ],
    }
]

# Preparation for inference
text = processor.apply_chat_template(
    messages, tokenize=False, add_generation_prompt=True
)
image_inputs, video_inputs = process_vision_info(messages)
inputs = processor(
    text=[text],
    images=image_inputs,
    videos=video_inputs,
    padding=True,
    return_tensors="pt",
)
inputs = inputs.to("cuda")

# Inference: Generation of the output
generated_ids = model.generate(**inputs, max_new_tokens=128)
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)

Buf

    buffer = ""
    for new_text in streamer:
        buffer += new_text
        # Remove <|im_end|> or similar tokens from the output
        buffer = buffer.replace("<|im_end|>", "")
        yield buffer

Key Features

1. Vision-Language Integration:

   • Combines image understanding with natural language processing to convert images into text.

2. Optical Character Recognition (OCR):

   • Extracts and processes textual information from images with high accuracy.

3. Math and LaTeX Support:

   • Solves math problems and outputs equations in LaTeX format.

4. Conversational Capabilities:

   • Designed to handle multi-turn interactions, providing context-aware responses.

5. Image-Text-to-Text Generation:

   • Inputs can include images, text, or a combination, and the model generates descriptive or problem-solving text.

6. Secure Weight Format:

   • Uses Safetensors for faster and more secure model weight loading.

---

Training Details

• Base Model: Qwen/Qwen2-VL-2B-Instruct

• Model Size:

  • 2.21 Billion parameters
  • Optimized for BF16 tensor type, enabling efficient inference.

• Specializations:

  • OCR tasks in images containing text.
  • Mathematical reasoning and LaTeX output for equations.

---