Create app.py

app.py

@@ -0,0 +1,49 @@
+import gradio as gr
+import torch
+from transformers import AutoTokenizer, T5ForConditionalGeneration
+
+# 1. Define your fine-tuned model
+model_name = "Spark2scale/Spark2Scale"
+
+# 2. Load the model and tokenizer
+# Using bfloat16 is crucial here to fit the 3B model into the 16GB free tier
+print("Loading tokenizer...")
+tokenizer = AutoTokenizer.from_pretrained(model_name)
+
+print("Loading model in bfloat16...")
+model = T5ForConditionalGeneration.from_pretrained(
+    model_name,
+    torch_dtype=torch.bfloat16
+)
+model.eval() # Set to evaluation mode
+
+# 3. Define the prediction function
+def evaluate_idea(startup_idea):
+    # Tokenize input
+    inputs = tokenizer(startup_idea, return_tensors="pt")
+    
+    # Generate text (adjust max_new_tokens based on your desired report length)
+    with torch.no_grad():
+        outputs = model.generate(
+            **inputs,
+            max_new_tokens=512,
+            temperature=0.7,
+            do_sample=True
+        )
+        
+    # Decode and return the result
+    result = tokenizer.decode(outputs[0], skip_special_tokens=True)
+    return result
+
+# 4. Build the Gradio Interface
+# This UI automatically generates the REST API endpoint
+iface = gr.Interface(
+    fn=evaluate_idea,
+    inputs=gr.Textbox(lines=5, placeholder="Enter the startup idea here...", label="Startup Idea"),
+    outputs=gr.Textbox(label="Evaluation Report"),
+    title="Spark2Scale Idea Evaluator",
+    description="Backend API for the Spark2Scale evaluation agent."
+)
+
+# 5. Launch the app
+iface.launch()