File size: 2,698 Bytes
e451d68 97aeb66 e451d68 97aeb66 043eee6 97aeb66 a45558b |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 |
---
license: apache-2.0
base_model:
- Qwen/Qwen2.5-VL-7B-Instruct
---
### Model Sources
- **Repository:** https://github.com/maifoundations/DualMindVLM
- **Paper:** https://arxiv.org/pdf/2511.16670
### Quick Start
The model is fine-tuned from Qwen2.5-VL-7B-Instruct. We provide an inference example using the Transformers inference backend.
```
import torch
from transformers import Qwen2_5_VLForConditionalGeneration, AutoTokenizer, AutoProcessor
from qwen_vl_utils import process_vision_info
model = Qwen2_5_VLForConditionalGeneration.from_pretrained(
"maifoundations/DualMindVLM",
torch_dtype=torch.bfloat16,
attn_implementation="flash_attention_2",
device_map="auto",
)
# default processer
processor = AutoProcessor.from_pretrained("maifoundations/DualMindVLM")
SYSTEM_PROMPT = """You are a Vision-Language Model answering questions about images.
Follow these rules strictly:
1. Judge the length of reasoning needed.
- Short: start with "Short Thinking:".
- Long: start with "Long Thinking:".
2. Short Thinking: give a concise thinking process which is sufficient to answer the question, then provide the final answer.
3. Long Thinking: give a structured reasoning process of the question and the image, including question analysis, visual details description, self-verification and then provide the final answer.
4. The final answer MUST BE put in \\boxed{}."""
messages = [
{"role": "system", "content": [{"type": "text", "text": SYSTEM_PROMPT}]},
{
"role": "user",
"content": [
{
"type": "image",
"image": image_path,
},
{"type": "text", "text": question},
],
}
]
# 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=1024)
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)
```
### Citation
```bibtex
@article{lin2025dualmindvlm,
title = {Learning to Think Fast and Slow for Visual Language Models},
author = {Chenyu Lin and Cheng Chi and Jinlin Wu and Sharon Li and Kaiyang Zhou},
journal = {arXiv preprint arXiv:2511.16670},
year = {2025}
}
|