Qika/DeepMedix-R1

Model Usage:

from transformers import Qwen2_5_VLForConditionalGeneration, AutoTokenizer, AutoProcessor
from qwen_vl_utils import process_vision_info

model = Qwen2_5_VLForConditionalGeneration.from_pretrained(
    model_path,
    torch_dtype=torch.bfloat16,
    attn_implementation="flash_attention_2",
    device_map="auto",
)
processor = AutoProcessor.from_pretrained(model_path, max_pixels=262144)


reason_prompt = r"You FIRST think about the reasoning process as an internal monologue and then provide the final answer. The reasoning process MUST BE enclosed within <think> </think> tags. During this reasoning process, prioritize analyzing the local regions of the image by leveraging the bounding box coordinates in the format [x_min, y_min, x_max, y_max]. The final answer MUST BE put in \boxed{}. An example is like: <think> reasoning process 1 with [x_min1, y_min1, x_max1, y_max1]; reasoning process 2 with [x_min2, y_min2, x_max2, y_max2] </think>. The answer is: \boxed{answer}."

def get_label(images, content1):
    content_list = []
    for image_url in images:
        content_list.append({
            "type": "image",
            "image": image_url,
        })
    if mode == 'think':
        content_list.append({"type": "text",
                             "text": content1 + '\n' + reason_prompt + '\n'})
    else:
        content_list.append({"type": "text",
                             "text": content1})
    messages = [
        {
            "role": "user",
            "content": content_list
        }
    ]

    # Preparation for inference
    text = processor.apply_chat_template(
        messages, tokenize=False, add_generation_prompt=True
    )
    # print(text)
    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=4096, do_sample=True, temperature=0.6)
    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)
    # print(output_text[0])
    return output_text[0]