add simple side-by-side

fastchat/serve/gradio_block_arena_txt2img.py

@@ -3,18 +3,29 @@
 import requests
 import base64
 import io
+import hashlib
 from PIL import Image
+import datetime
+import json
+
+from fastchat.constants import LOGDIR
+from fastchat.utils import upload_image_file_to_gcs
 
 FIREWORKS_API_KEY = os.getenv("FIREWORKS_API_KEY")
-API_BASE = "https://api.fireworks.ai/inference/v1/workflows/accounts/fireworks/models/{}/text_to_image"
+API_BASE = "https://api.fireworks.ai/inference/v1/workflows/accounts/fireworks/models/{model}/text_to_image"
 DUMMY_MODELS = ["stable-diffusion-3p5-medium", 
                 "stable-diffusion-3p5-large",
                 "stable-diffusion-3p5-large-turbo",
                 "flux-1-dev-fp8",
                 "flux-1-schnell-fp8",
                 ]
 
-def generate_image(prompt, model):
+def get_conv_log_filename():
+    t = datetime.datetime.now()
+    conv_log_filename = f"{t.year}-{t.month:02d}-{t.day:02d}-conv.json"
+    return os.path.join(LOGDIR, f"txt2img-{conv_log_filename}")
+
+def generate_image(model, prompt):
     """Generate image from text prompt using Fireworks API"""
     headers = {
         "Authorization": f"Bearer {FIREWORKS_API_KEY}",
@@ -38,29 +49,65 @@ def generate_image(prompt, model):
 
         image_bytes = response.content
         image = Image.open(io.BytesIO(image_bytes))
-
-        return image
-
+
     except requests.exceptions.RequestException as e:
         return f"Error generating image: {str(e)}"
 
+    log_filename = get_conv_log_filename()
+    image_hash = hashlib.md5(image.tobytes()).hexdigest()
+    image_filename = f"{image_hash}.png"
+    upload_image_file_to_gcs(image, image_filename)
+
+    with open(log_filename, "a") as f:
+        data = {
+            "model": model,
+            "prompt": prompt,
+            "image_filename": image_filename
+        }
+        f.write(json.dumps(data) + "\n")
+
+    return image
+
+def generate_image_multi(model_left, model_right, prompt):
+    images = []
+    for model in [model_left, model_right]:
+        images.append(generate_image(model, prompt))
+
+    return images
+
+
 # Create Gradio interface
 with gr.Blocks(title="Text to Image Generator") as demo:
     gr.Markdown("# Text to Image Generator")
     gr.Markdown("Enter a text prompt to generate an image")
 
+
+    num_sides = 2
+    model_selectors = [None] * num_sides
+
     with gr.Column():        
         with gr.Group():
-            model_selector = gr.Dropdown(
-                choices=DUMMY_MODELS,
-                interactive=True,
-                show_label=False,
-                container=False,
-            )
-            image_output = gr.Image(
-                label="Generated Image",
-                type="pil"
-            )
+            with gr.Row():
+                for i in range(num_sides):
+                    model_selectors[i] = gr.Dropdown(
+                        choices=DUMMY_MODELS,
+                        value=DUMMY_MODELS[i] if DUMMY_MODELS else "",
+                        interactive=True,
+                        show_label=False,
+                        container=False,
+                    )
+
+
+            with gr.Group():
+                with gr.Row():
+                    output_left = gr.Image(
+                        type="pil",
+                        show_label=False
+                    )
+                    output_right = gr.Image(
+                        type="pil",
+                        show_label=False
+                    )
 
         with gr.Row():
             text_input = gr.Textbox(
@@ -72,15 +119,15 @@ def generate_image(prompt, model):
 
     # Handle generation
     send_btn.click(
-        fn=generate_image,
-        inputs=[text_input, model_selector],
-        outputs=image_output
+        fn=generate_image_multi,
+        inputs=model_selectors + [text_input],
+        outputs=[output_left, output_right]
     )
 
     text_input.submit(
-        fn=generate_image,
-        inputs=[text_input, model_selector],
-        outputs=image_output
+        fn=generate_image_multi,
+        inputs=model_selectors + [text_input],
+        outputs=[output_left, output_right]
     )
 
 if __name__ == "__main__":