Add Nucleus-Image generation app with grouped_mm polyfill

Gradio app for NucleusAI/Nucleus-Image on ZeroGPU with text KV cache support.
Includes polyfill for torch.nn.functional.grouped_mm using the aten kernel.

app.py

@@ -6,6 +6,12 @@ import spaces
 import torch
 from diffusers import DiffusionPipeline
 
+# Polyfill: expose torch.ops.aten._grouped_mm as F.grouped_mm if not already present
+if not hasattr(torch.nn.functional, "grouped_mm"):
+    def _grouped_mm(input, mat2, *, offs=None, bias=None, out_dtype=None):
+        return torch.ops.aten._grouped_mm(input, mat2, offs=offs, bias=bias, out_dtype=out_dtype)
+    torch.nn.functional.grouped_mm = _grouped_mm
+
 MODEL_NAME = "NucleusAI/Nucleus-Image"
 MAX_SEED = np.iinfo(np.int32).max
 
@@ -89,14 +95,13 @@ with gr.Blocks(css=CSS, theme=gr.themes.Soft()) as demo:
             """
         )
 
-        with gr.Row():
-            prompt = gr.Textbox(
-                label="Prompt",
-                placeholder="Describe the image you want to generate...",
-                lines=3,
-                scale=4,
-            )
-            run_btn = gr.Button("Generate", variant="primary", scale=1)
+        prompt = gr.Textbox(
+            label="Prompt",
+            placeholder="Describe the image you want to generate...",
+            lines=3,
+            scale=4,
+        )
+        run_btn = gr.Button("Generate", variant="primary", scale=1)
 
         result = gr.Image(label="Result", show_label=False, format="png")
 
@@ -111,7 +116,7 @@ with gr.Blocks(css=CSS, theme=gr.themes.Soft()) as demo:
                     label="Inference Steps", minimum=10, maximum=80, step=1, value=50
                 )
                 guidance_scale = gr.Slider(
-                    label="Guidance Scale", minimum=1.0, maximum=15.0, step=0.5, value=8.0
+                    label="Guidance Scale", minimum=1.0, maximum=15.0, step=0.5, value=3.5
                 )
             with gr.Row():
                 seed = gr.Slider(