Views
No views yet
1from transformers import Qwen2VLForConditionalGeneration, AutoProcessor
2from qwen_vl_utils import process_vision_info
3import torch
4
5# Load model and processor
6model = Qwen2VLForConditionalGeneration.from_pretrained(
7 "v1v1d1/vivid_docmatix_gemma3_4b_en_kn_hi_10k_50",
8 torch_dtype=torch.bfloat16,
9 device_map="auto"
10)
11processor = AutoProcessor.from_pretrained("v1v1d1/vivid_docmatix_gemma3_4b_en_kn_hi_10k_50")
12
13# Prepare inputs
14messages = [
15 {
16 "role": "user",
17 "content": [
18 {"type": "image", "image": "path/to/image.jpg"},
19 {"type": "text", "text": "Extract all text from this document in Kannada."},
20 ],
21 }
22]
23
24text = processor.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
25image_inputs, video_inputs = process_vision_info(messages)
26inputs = processor(
27 text=[text],
28 images=image_inputs,
29 videos=video_inputs,
30 padding=True,
31 return_tensors="pt",
32).to(model.device)
33
34# Generate
35with torch.no_grad():
36 output_ids = model.generate(**inputs, max_new_tokens=4096)
37 generated_ids = [
38 output_ids[len(input_ids):]
39 for input_ids, output_ids in zip(inputs.input_ids, output_ids)
40 ]
41 output_text = processor.batch_decode(
42 generated_ids, skip_special_tokens=True, clean_up_tokenization_spaces=True
43 )[0]
44
45print(output_text)1@misc{nayana-vivid-docmatix,
2 title={VIVID Docmatix: Multilingual Vision-Language Model},
3 author={Nayana Team},
4 year={2026},
5 url={https://huggingface.co/v1v1d1/vivid_docmatix_gemma3_4b_en_kn_hi_10k_50}
6}