Spaces:
Sleeping
Sleeping
Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,68 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import gradio as gr
|
2 |
+
from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
|
3 |
+
from diffusers import StableDiffusionPipeline
|
4 |
+
from sentence_transformers import SentenceTransformer, util
|
5 |
+
import torch
|
6 |
+
|
7 |
+
# --- Load models ---
|
8 |
+
device = "cuda" if torch.cuda.is_available() else "cpu"
|
9 |
+
|
10 |
+
# Text-to-text model
|
11 |
+
text_model_name = "google/flan-t5-small"
|
12 |
+
text_tokenizer = AutoTokenizer.from_pretrained(text_model_name)
|
13 |
+
text_model = AutoModelForSeq2SeqLM.from_pretrained(text_model_name)
|
14 |
+
|
15 |
+
# Text-to-image model
|
16 |
+
image_model_id = "runwayml/stable-diffusion-v1-5"
|
17 |
+
image_pipe = StableDiffusionPipeline.from_pretrained(
|
18 |
+
image_model_id, torch_dtype=torch.float16 if device=="cuda" else torch.float32
|
19 |
+
)
|
20 |
+
image_pipe = image_pipe.to(device)
|
21 |
+
|
22 |
+
# Sentence similarity model
|
23 |
+
embedder = SentenceTransformer('all-MiniLM-L6-v2')
|
24 |
+
|
25 |
+
# Example prompts that mean “generate image”
|
26 |
+
image_triggers = [
|
27 |
+
"generate an image of",
|
28 |
+
"draw a",
|
29 |
+
"create a picture of",
|
30 |
+
"show me a",
|
31 |
+
"visualize",
|
32 |
+
"render",
|
33 |
+
"sketch",
|
34 |
+
]
|
35 |
+
|
36 |
+
# --- Main Logic ---
|
37 |
+
def multimodal_agent(prompt):
|
38 |
+
# Step 1: Check similarity to image prompts
|
39 |
+
prompt_embedding = embedder.encode(prompt, convert_to_tensor=True)
|
40 |
+
trigger_embeddings = embedder.encode(image_triggers, convert_to_tensor=True)
|
41 |
+
|
42 |
+
cosine_scores = util.pytorch_cos_sim(prompt_embedding, trigger_embeddings)
|
43 |
+
max_score = torch.max(cosine_scores).item()
|
44 |
+
|
45 |
+
# Step 2: Threshold to decide
|
46 |
+
if max_score > 0.65:
|
47 |
+
# Generate image
|
48 |
+
image = image_pipe(prompt).images[0]
|
49 |
+
return None, image # Return image only
|
50 |
+
else:
|
51 |
+
# Generate text
|
52 |
+
inputs = text_tokenizer(prompt, return_tensors="pt")
|
53 |
+
outputs = text_model.generate(**inputs, max_new_tokens=100)
|
54 |
+
text = text_tokenizer.decode(outputs[0], skip_special_tokens=True)
|
55 |
+
return text, None # Return text only
|
56 |
+
|
57 |
+
# --- Gradio UI ---
|
58 |
+
with gr.Blocks() as demo:
|
59 |
+
gr.Markdown("# 🤖 Smart Multimodal AI Agent\nDecides text or image based on your prompt.")
|
60 |
+
|
61 |
+
input_box = gr.Textbox(label="Enter your prompt")
|
62 |
+
output_text = gr.Textbox(label="Text Output")
|
63 |
+
output_image = gr.Image(label="Image Output")
|
64 |
+
|
65 |
+
btn = gr.Button("Generate")
|
66 |
+
btn.click(multimodal_agent, inputs=input_box, outputs=[output_text, output_image])
|
67 |
+
|
68 |
+
demo.launch()
|