app.py

import gradio as gr
from transformers import AutoTokenizer

tokenizer = AutoTokenizer.from_pretrained("HuggingFaceH4/zephyr-7b-beta")

demo_conversation1 = [
    {"role": "user", "content": "Hi there!"},
    {"role": "assistant", "content": "Hello, human!"}
]

demo_conversation2 = [
    {"role": "system", "content": "You are a helpful chatbot."},
    {"role": "user", "content": "Hi there!"}
]

conversations = [demo_conversation1, demo_conversation2]

def apply_chat_template(template):
    tokenizer.chat_template = template
    outputs = []
    for i, conversation in enumerate(conversations):
        without_gen = tokenizer.apply_chat_template(conversation, tokenize=False)
        with_gen = tokenizer.apply_chat_template(conversation, tokenize=False, add_generation_prompt=True)
        out = f"Conversation {i} without generation prompt:\n\n{without_gen}\n\nConversation {i} with generation prompt:\n\n{with_gen}\n\n"
        outputs.append(out)
    return tuple(outputs)

iface = gr.Interface(fn=apply_chat_template, inputs="text", outputs=["text"] * len(conversations))
iface.launch()