add header and change theem

app.py

@@ -4,6 +4,20 @@ import spaces
 from threading import Thread
 from typing import Iterator
 
+# Add markdown header
+header = """
+# 🐦‍⬛ MagpieLMs: Open LLMs with Fully Transparent Alignment Recipes
+
+💬 We've aligned Llama-3.1-8B and a 4B version (distilled by NVIDIA) using purely synthetic data generated by our [Magpie](https://arxiv.org/abs/2406.08464) method. Our open-source post-training recipe includes: SFT and DPO data, all training configs + logs. This allows everyone to reproduce the alignment process for their own research. Note that our data does not contain any GPT-generated data, and has a much friendly license for both commercial and academic use.
+
+- **Magpie Collection**: [Magpie on Hugging Face](https://lnkd.in/g_pgX5Y2)
+- **Magpie Paper**: [Read the research paper](https://arxiv.org/abs/2406.08464) 
+
+Contact: [Zhangchen Xu](https://zhangchenxu.com) and [Bill Yuchen Lin](https://yuchenlin.xyz).
+
+---
+"""
+
 # Load model and tokenizer
 model_name = "Magpie-Align/MagpieLM-4B-Chat-v0.1"
 
@@ -11,7 +25,8 @@ device = "cuda" # the device to load the model onto
 tokenizer = AutoTokenizer.from_pretrained(model_name, trust_remote_code=True)
 model = AutoModelForCausalLM.from_pretrained(
     model_name,
-    torch_dtype="auto"
+    torch_dtype="auto",
+    ignore_mismatched_sizes=True
 )
 model.to(device)
 
@@ -64,21 +79,23 @@ def respond(
 demo = gr.ChatInterface(
     respond,
     additional_inputs=[
-        gr.Textbox(value="You are Magpie, a helpful AI assistant.", label="System message"),
-        gr.Slider(minimum=1, maximum=2048, value=512, step=1, label="Max new tokens"),
-        gr.Slider(minimum=0.1, maximum=4.0, value=0.6, step=0.1, label="Temperature"),
+        gr.Textbox(value="You are Magpie, a helpful AI assistant. For simple qeuries, try to answer them directly; for complex questions, try to think step-by-step before providing an answer.", label="System message"),
+        gr.Slider(minimum=128, maximum=2048, value=512, step=64, label="Max new tokens"),
+        gr.Slider(minimum=0.1, maximum=4.0, value=0.7, step=0.1, label="Temperature"),
         gr.Slider(
             minimum=0.1,
             maximum=1.0,
             value=0.9,
-            step=0.05,
+            step=0.1,
             label="Top-p (nucleus sampling)",
         ),
         gr.Slider(minimum=0.5, maximum=1.5, value=1.0, step=0.1, label="Repetation Penalty"),
     ],
+    description=header,  # Add the header as the description
+    title="MagpieLM-4B Chat (v0.1)",
+    theme=gr.themes.Soft()
 )
 
-
 if __name__ == "__main__":
     demo.queue()
     demo.launch(share=True)