|
import spaces |
|
import torch |
|
import re |
|
import gradio as gr |
|
from transformers import AutoTokenizer, AutoModelForCausalLM |
|
from PIL import Image |
|
|
|
if torch.cuda.is_available(): |
|
device, dtype = "cuda", torch.float16 |
|
else: |
|
device, dtype = "cpu", torch.float32 |
|
|
|
model_id = "vikhyatk/moondream2" |
|
revision = "2024-04-02" |
|
tokenizer = AutoTokenizer.from_pretrained(model_id, revision=revision) |
|
moondream = AutoModelForCausalLM.from_pretrained( |
|
model_id, trust_remote_code=True, revision=revision, torch_dtype=dtype |
|
).to(device=device) |
|
moondream.eval() |
|
|
|
@spaces.GPU(duration=10) |
|
def answer_questions(image_tuples, prompt_text): |
|
result = "" |
|
|
|
|
|
prompts = [p.strip() for p in prompt_text.split(',')] |
|
|
|
|
|
image_embeds = [img[0] for img in image_tuples if img[0] is not None] |
|
|
|
|
|
if len(image_embeds) != len(prompts): |
|
return ("Error: The number of images input and prompts input (seperate by commas in input text field) must be the same.") |
|
|
|
answers = moondream.batch_answer( |
|
images=image_embeds, |
|
prompts=prompts, |
|
tokenizer=tokenizer, |
|
) |
|
|
|
for question, answer in zip(prompts, answers): |
|
|
|
|
|
|
|
result += (f"Q: {question}\nA: {answer}\n\n") |
|
|
|
return result |
|
|
|
with gr.Blocks() as demo: |
|
gr.Markdown("# moondream2 unofficial batch processing demo") |
|
gr.Markdown("1. Select images\n2. Enter prompts (one prompt for each image provided) separated by commas. Ex: Describe this image, What is in this image?\n\n") |
|
gr.Markdown("*Tested and Running on free CPU space tier currently so results may take a bit to process compared to using GPU space hardware*") |
|
gr.Markdown("## π moondream2\nA tiny vision language model. [GitHub](https://github.com/vikhyatk/moondream)") |
|
with gr.Row(): |
|
img = gr.Gallery(label="Upload Images", type="pil") |
|
prompt = gr.Textbox(label="Input Prompts", placeholder="Enter prompts (one prompt for each image provided) separated by commas. Ex: Describe this image, What is in this image?", lines=8) |
|
submit = gr.Button("Submit") |
|
output = gr.TextArea(label="Responses", lines=8) |
|
submit.click(answer_questions, [img, prompt], output) |
|
|
|
demo.queue().launch() |
|
|