Create app.py

app.py

@@ -0,0 +1,44 @@
+import gradio as gr
+from transformers import AutoTokenizer, AutoModelForCausalLM
+
+# Load your model (using GPT-2 as an example)
+model_name = "gpt2"
+tokenizer = AutoTokenizer.from_pretrained(model_name)
+model = AutoModelForCausalLM.from_pretrained(model_name)
+
+def generate_completions(prompt):
+    # Define decoding strategies with corresponding parameters
+    strategies = {
+        "Greedy": {"do_sample": False},
+        "Beam Search": {"num_beams": 5, "early_stopping": True},
+        "Top-k Sampling": {"do_sample": True, "top_k": 50},
+        "Top-p Sampling": {"do_sample": True, "top_p": 0.95}
+    }
+    
+    results = {}
+    input_ids = tokenizer.encode(prompt, return_tensors="pt")
+    
+    for strategy, params in strategies.items():
+        # Generate output using the specific strategy
+        output_ids = model.generate(input_ids, max_length=50, **params)
+        output_text = tokenizer.decode(output_ids[0], skip_special_tokens=True)
+        results[strategy] = output_text
+    
+    return results["Greedy"], results["Beam Search"], results["Top-k Sampling"], results["Top-p Sampling"]
+
+# Define the Gradio interface
+interface = gr.Interface(
+    fn=generate_completions,
+    inputs=gr.inputs.Textbox(lines=3, placeholder="Enter your prompt here...", label="Prompt"),
+    outputs=[
+        gr.outputs.Textbox(label="Greedy"),
+        gr.outputs.Textbox(label="Beam Search"),
+        gr.outputs.Textbox(label="Top-k Sampling"),
+        gr.outputs.Textbox(label="Top-p Sampling"),
+    ],
+    title="LLM Decoding Strategies Comparison",
+    description="Enter a prompt to see how different decoding strategies affect the output of a language model."
+)
+
+if __name__ == "__main__":
+    interface.launch()