app.py

app.py

@@ -0,0 +1,57 @@
+import gradio as gr
+import torch
+from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
+
+# 1. Load a model that fits in the free tier
+# 'flan-t5-large' is powerful but small enough for the free CPU tier
+model_name = 'google/flan-t5-large'
+
+print("Loading model... this may take a minute.")
+tokenizer = AutoTokenizer.from_pretrained(model_name)
+model = AutoModelForSeq2SeqLM.from_pretrained(model_name)
+
+# Use GPU if available, otherwise CPU
+device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
+model = model.to(device)
+
+# 2. Define the generation function
+def generate_text(task_prefix, input_text):
+    # Dynamic prompt construction
+    prompt = f"{task_prefix}: {input_text}"
+    
+    inputs = tokenizer(prompt, return_tensors="pt").to(device)
+    
+    # Generate output
+    output_ids = model.generate(
+        **inputs,
+        max_length=64,
+        num_beams=4,
+        early_stopping=True,
+        no_repeat_ngram_size=2
+    )
+    
+    return tokenizer.decode(output_ids[0], skip_special_tokens=True)
+
+# 3. Create the Web Interface
+# We map the choices to the actual prefixes the model understands
+task_choices = [
+    "summarize", 
+    "translate French to English", 
+    "paraphrase", 
+    "generate question", 
+    "sst2 sentence"
+]
+
+demo = gr.Interface(
+    fn=generate_text,
+    inputs=[
+        gr.Dropdown(choices=task_choices, label="Select Task", value="summarize"),
+        gr.Textbox(label="Input Text", placeholder="Enter your text here...")
+    ],
+    outputs=gr.Textbox(label="AI Output"),
+    title="Multi-Task AI Generator",
+    description="Select a task and enter text. Powered by Google Flan-T5."
+)
+
+# 4. Launch
+demo.launch()