Update app.py

app.py

@@ -51,13 +51,28 @@ def generate_pairs(context):
         output += f"## Instruction {index + 1}:\n{pair['Q']}\n## Response {index + 1}:\n{pair['A']}\n\n"
     return output
 
+description = """
+## Instruction Pre-Training: Language Models as Supervised Multitask Learners
+
+This demo implements the instruction synthesis approach from the paper ["Instruction Pre-Training: Language Models are Supervised Multitask Learners"](https://huggingface.co/papers/2406.14491).
+
+### Method:
+1. An instruction synthesizer is trained on diverse datasets to generate instruction-response pairs from raw text.
+2. The synthesizer augments raw pre-training corpora with synthesized instruction-response pairs.
+3. Language models are then pre-trained on this augmented data, combining unsupervised and supervised multitask learning.
+
+This approach enhances model performance and generalization, particularly benefiting from further instruction tuning.
+
+Try it out by entering some text below!
+"""
+
 # Create Gradio interface
 iface = gr.Interface(
     fn=generate_pairs,
     inputs=gr.Textbox(lines=5, label="Enter context here"),
     outputs=gr.Textbox(lines=20, label="Generated Instruction-Response Pairs"),
     title="Instruction-Response Pair Generator",
-    description="Enter a context, and the model will generate relevant instruction-response pairs."
+    description=description
 )
 
 # Launch the interface