Image-to-Text
Transformers
Safetensors
qwen3_5
image-text-to-text
vision-language
vlm
document-understanding
structured-extraction
information-extraction
ocr
document-to-markdown
markdown
rag
reasoning
multilingual
conversational
Instructions to use numind/NuExtract3 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use numind/NuExtract3 with Transformers:
# Use a pipeline as a high-level helper # Warning: Pipeline type "image-to-text" is no longer supported in transformers v5. # You must load the model directly (see below) or downgrade to v4.x with: # 'pip install "transformers<5.0.0' from transformers import pipeline pipe = pipeline("image-to-text", model="numind/NuExtract3") messages = [ { "role": "user", "content": [ {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/p-blog/candy.JPG"}, {"type": "text", "text": "What animal is on the candy?"} ] }, ] pipe(text=messages)# Load model directly from transformers import AutoProcessor, AutoModelForImageTextToText processor = AutoProcessor.from_pretrained("numind/NuExtract3") model = AutoModelForImageTextToText.from_pretrained("numind/NuExtract3") messages = [ { "role": "user", "content": [ {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/p-blog/candy.JPG"}, {"type": "text", "text": "What animal is on the candy?"} ] }, ] inputs = processor.apply_chat_template( messages, add_generation_prompt=True, tokenize=True, return_dict=True, return_tensors="pt", ).to(model.device) outputs = model.generate(**inputs, max_new_tokens=40) print(processor.decode(outputs[0][inputs["input_ids"].shape[-1]:])) - Notebooks
- Google Colab
- Kaggle
Update README.md
Browse files
README.md
CHANGED
|
@@ -99,36 +99,6 @@ To add other evaluate references, we used our structured extraction benchmark to
|
|
| 99 |
|
| 100 |
<img src="md2st.svg" width="1000"/>
|
| 101 |
|
| 102 |
-
#### No thinking
|
| 103 |
-
|
| 104 |
-
| Model | Average score | Avg. num tokens answer |
|
| 105 |
-
| ---------------------| ------------: | ---------------------: |
|
| 106 |
-
| *Generalist models* | | |
|
| 107 |
-
| NuExtract3 |**0.683 ± 0.021**|1821 |
|
| 108 |
-
| Qwen3.5-4B |0.604 ± 0.025 |1797 |
|
| 109 |
-
| gemma-4-E4B |0.502 ± 0.027 |613 |
|
| 110 |
-
| GLM-4.6V-Flash |0.579 ± 0.029 |928 |
|
| 111 |
-
| Nemotron-3-Nano-Omni |0.640 ± 0.024 |1382 |
|
| 112 |
-
| granite-vision-4.1-4b|0.468 ± 0.026 |750 |
|
| 113 |
-
| Ministral-3-3B |0.521 ± 0.025 |3248 |
|
| 114 |
-
| *Specialized OCR/Markdown models* | | |
|
| 115 |
-
| GLM-OCR |0.631 ± 0.024 |1247 |
|
| 116 |
-
| LightOnOCR-2-1B |0.633 ± 0.024 |1073 |
|
| 117 |
-
| olmOCR-2-7B-1025 |0.587 ± 0.024 |732 |
|
| 118 |
-
| chandra-ocr-2 |0.665 ± 0.021 |2012 |
|
| 119 |
-
| PaddleOCR-VL-1.5 |0.433 ± 0.025 |747 |
|
| 120 |
-
|
| 121 |
-
#### Thinking
|
| 122 |
-
| Model | Average score | Avg. num tokens thinking | Avg. num tokens answer |
|
| 123 |
-
| -------------------- | ----------------: | -----------------------: | ---------------------: |
|
| 124 |
-
| NuExtract3 | **0.701 ± 0.019** | 338 | 1981 |
|
| 125 |
-
| Qwen3.5-4B | 0.662 ± 0.022 | 6552 | 1547 |
|
| 126 |
-
| gemma-4-E4B-it | 0.550 ± 0.023 | 790 | 672 |
|
| 127 |
-
| GLM-4.6V-Flash | 0.638 ± 0.023 | 1973 | 886 |
|
| 128 |
-
| Nemotron-3-Nano-Omni | 0.626 ± 0.025 | 11725 | 1040 |
|
| 129 |
-
| Ministral-3-3B | 0.542 ± 0.026 | 7593 | 773|
|
| 130 |
-
|
| 131 |
-
|
| 132 |
|
| 133 |
# Using NuExtract
|
| 134 |
|
|
|
|
| 99 |
|
| 100 |
<img src="md2st.svg" width="1000"/>
|
| 101 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 102 |
|
| 103 |
# Using NuExtract
|
| 104 |
|