|
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 |
|
without_gen = tokenizer.apply_chat_template(conversation, tokenize=False) |
|
with_gen = tokenizer.apply_chat_template(conversation, tokenize=False, add_generation_prompt=True) |
|
return f"Without generation prompt:\n\n{without_gen}\n\nWith generation prompt:\n\n{with_gen}" |
|
|
|
iface = gr.Interface(fn=apply_chat_template, inputs="text", outputs="text") |
|
iface.launch() |
|
|