Setup Gradio app with Unsloth

app.py

@@ -0,0 +1,35 @@
+import gradio as gr
+from unsloth import FastVisionModel
+from PIL import Image
+import torch
+
+device = "cuda" if torch.cuda.is_available() else "cpu"
+
+# Load the Unsloth FastVisionModel
+model, tokenizer = FastVisionModel.from_pretrained(
+    model_name = "unsloth/mllama-vision",
+    max_seq_length = 2048,
+    dtype = torch.float16,
+    load_in_4bit = True,
+).to(device)
+
+def analyze_image(image, prompt):
+    inputs = tokenizer(
+        image, 
+        prompt, 
+        return_tensors="pt"
+    ).to(device)
+
+    outputs = model.generate(**inputs, max_new_tokens=100)
+    result = tokenizer.decode(outputs[0], skip_special_tokens=True)
+    return result
+
+iface = gr.Interface(
+    fn=analyze_image,
+    inputs=[gr.Image(type="pil"), "text"],
+    outputs="text",
+    title="FloorPlan Vision AI",
+    description="Upload an image and enter your prompt for AI analysis."
+)
+
+iface.launch()

requirements.txt

@@ -0,0 +1,6 @@
+unsloth
+torch
+transformers
+sentencepiece
+datasets
+Pillow