Create app.py

app.py

@@ -0,0 +1,105 @@
+import gradio as gr
+import torch
+from transformers import GPT2LMHeadModel, GPT2Tokenizer
+
+
+def load_model():
+    # 'gpt2' is lightweight and works great on Hugging Face's free tier
+    model_name = "gpt2"
+
+    tokenizer = GPT2Tokenizer.from_pretrained(model_name)
+    model = GPT2LMHeadModel.from_pretrained(model_name)
+
+    tokenizer.pad_token = tokenizer.eos_token
+
+    # Hugging Face Spaces will automatically detect the runtime environment
+    device = "cuda" if torch.cuda.is_available() else "cpu"
+    model = model.to(device)
+
+    return model, tokenizer, device
+
+
+# Initialize model globally
+model, tokenizer, device = load_model()
+
+
+def generate_text(prompt, max_length, num_return_sequences):
+    if not prompt.strip():
+        return "Please enter a valid prompt."
+
+    inputs = tokenizer(prompt, return_tensors="pt").to(device)
+
+    with torch.no_grad():
+        output_sequences = model.generate(
+            input_ids=inputs["input_ids"],
+            attention_mask=inputs["attention_mask"],
+            max_length=int(max_length),
+            num_return_sequences=int(num_return_sequences),
+            do_sample=True,  # Crucial for returning multiple unique sequences
+            temperature=0.7,
+            top_k=50,
+            top_p=0.95,
+            no_repeat_ngram_size=2,
+        )
+
+    # Formats multiple outputs cleanly inside the single text box
+    final_output = ""
+    for i, seq in enumerate(output_sequences):
+        decoded_text = tokenizer.decode(seq, skip_special_tokens=True)
+        final_output += f"--- Sequence {i+1} ---\n{decoded_text}\n\n"
+
+    return final_output
+
+
+# --- Gradio UI Layout ---
+with gr.Blocks(theme=gr.themes.Soft()) as demo:
+    gr.Markdown("# 🤖 GPT-2 Text Generator")
+    gr.Markdown(
+        "Generate text using a pre-trained GPT-2 model. Enter a prompt, choose generation settings, and see the AI complete your text!"
+    )
+
+    with gr.Row():
+        with gr.Column(scale=1):
+            prompt_input = gr.Textbox(
+                label="Prompt",
+                placeholder="About Marvel Avengers...",
+                lines=4,
+            )
+            max_length_slider = gr.Slider(
+                minimum=20,
+                maximum=200,
+                value=100,
+                step=10,
+                label="Max Length of Generated Text",
+            )
+            sequences_slider = gr.Slider(
+                minimum=1,
+                maximum=5,
+                value=2,
+                step=1,
+                label="Number of Sequences to Return",
+            )
+
+            with gr.Row():
+                clear_btn = gr.Button("Clear")
+                submit_btn = gr.Button("Submit", variant="primary")
+
+        with gr.Column(scale=1):
+            output_text = gr.Textbox(
+                label="Generated Text", lines=12, interactive=False
+            )
+
+    # Click actions
+    submit_btn.click(
+        fn=generate_text,
+        inputs=[prompt_input, max_length_slider, sequences_slider],
+        outputs=output_text,
+    )
+
+    # Clear button action clears both fields
+    clear_btn.click(
+        fn=lambda: ("", ""), inputs=None, outputs=[prompt_input, output_text]
+    )
+
+# Launch without share=True (Hugging Face handles the web serving natively)
+demo.launch()