import gradio as gr
import requests
import base64
from PIL import Image
from io import BytesIO
def encode_image_to_base64(image: Image.Image) -> str:
buffered = BytesIO()
image.save(buffered, format="JPEG")
img_str = base64.b64encode(buffered.getvalue()).decode()
return f"data:image/jpeg;base64,{img_str}"
def query_vllm_api(image, temperature, max_tokens=12_000):
messages = []
if image is not None:
# Optional: Resize image if needed (to avoid huge uploads)
max_size = 1024
if max(image.size) > max_size:
ratio = max_size / max(image.size)
new_size = tuple(int(dim * ratio) for dim in image.size)
image = image.resize(new_size, Image.Resampling.LANCZOS)
image_b64 = encode_image_to_base64(image)
messages.append({
"role": "user",
"content": [
{"type": "image_url", "image_url": {"url": image_b64}}
]
})
payload = {
"model": "numind/NuMarkdown-8B-Thinking",
"messages": messages,
"max_tokens": max_tokens,
"temperature": temperature
}
try:
response = requests.post(
"http://localhost:8000/v1/chat/completions",
json=payload,
timeout=60
)
response.raise_for_status()
data = response.json()
result = data["choices"][0]["message"]["content"]
reasoning = result.split("")[1].split("")[0]
answer = result.split("")[1].split("")[0]
return reasoning, answer, answer
except requests.exceptions.RequestException as e:
return f"API request failed: {e}"
with gr.Blocks(title="NuMarkdown-8B-Thinking", theme=gr.themes.Soft()) as demo:
# Clean banner with centered content
gr.HTML("""
👁️ NuMarkdown-8B-Thinking
Upload an image to convert to Markdown!
NuMarkdown-8B-Thinking is the first reasoning OCR VLM. It is specifically trained to convert documents into clean Markdown files, well suited for RAG applications. It generates thinking tokens to figure out the layout of the document before generating the Markdown file. It is particularly good at understanding documents with weird layouts and complex tables.
NOTE: In this space we downsize large images and restrict the maximum output of the model, so performance could improve if you run the model yourself.
""")
with gr.Row():
with gr.Column():
temperature = gr.Slider(0.1, 1.5, value=0.6, step=0.1, label="Temperature")
img_in = gr.Image(type="pil", label="Upload Image")
btn = gr.Button("Generate Response")
with gr.Column():
thinking = gr.Textbox(label="Thinking Trace", lines=10)
raw_answer = gr.Textbox(label="Raw Output", lines=5)
output = gr.Markdown(label="Response")
btn.click(
query_vllm_api,
inputs=[img_in, temperature],
outputs=[thinking, raw_answer, output],
)
if __name__ == "__main__":
demo.launch(share=True)