Create app.py

app.py

@@ -0,0 +1,43 @@
+import gradio as gr
+from transformers import AutoTokenizer, AutoModelForCausalLM
+
+# Load the model and tokenizer
+model_name = "skylersterling/TopicGPT"
+tokenizer = AutoTokenizer.from_pretrained(model_name)
+model = AutoModelForCausalLM.from_pretrained(model_name)
+
+# Define the generation function
+def generate_text(context, max_tokens):
+    input_text = f"#CONTEXT# {context} #TOPIC#"
+    input_ids = tokenizer.encode(input_text, return_tensors='pt')
+    
+    # Generate tokens one by one
+    generated_ids = input_ids
+    for _ in range(max_tokens):
+        outputs = model(generated_ids)
+        next_token_id = outputs.logits[:, -1, :].argmax(dim=-1)
+        generated_ids = torch.cat([generated_ids, next_token_id.unsqueeze(0)], dim=1)
+        if next_token_id == tokenizer.encode("#TOPIC#", add_special_tokens=False)[0]:
+            break
+    
+    generated_text = tokenizer.decode(generated_ids[0], skip_special_tokens=True)
+    return generated_text
+
+# Create Gradio interface
+def gradio_interface():
+    context_input = gr.inputs.Textbox(lines=5, placeholder="Enter the context here...")
+    max_tokens_input = gr.inputs.Slider(minimum=1, maximum=200, default=50, step=1)
+    output_textbox = gr.outputs.Textbox()
+    
+    interface = gr.Interface(
+        fn=generate_text,
+        inputs=[context_input, max_tokens_input],
+        outputs=output_textbox,
+        title="TopicGPT Text Generation",
+        description="Generate text token-by-token using the TopicGPT model. The input should start with #CONTEXT# and end with #TOPIC#."
+    )
+    
+    interface.launch()
+
+if __name__ == "__main__":
+    gradio_interface()