|
|
|
import os |
|
from typing import Iterator |
|
|
|
import gradio as gr |
|
import spaces |
|
import torch |
|
from transformers import AutoModelForCausalLM, AutoTokenizer, TextIteratorStreamer |
|
|
|
from chat_interface_preference import ChatInterface |
|
|
|
MAX_MAX_NEW_TOKENS = 2048 |
|
DEFAULT_MAX_NEW_TOKENS = 1024 |
|
MAX_INPUT_TOKEN_LENGTH = int(os.getenv("MAX_INPUT_TOKEN_LENGTH", "8192")) |
|
|
|
if torch.cuda.is_available(): |
|
|
|
|
|
|
|
pass |
|
|
|
style = None |
|
|
|
AutoModelForCausalLM, AutoTokenizer, TextIteratorStreamer |
|
|
|
|
|
@spaces.GPU |
|
def generate( |
|
message: str, |
|
chat_history: list[tuple[str, str]], |
|
max_new_tokens: int = 1024, |
|
temperature: float = 0.06, |
|
top_p: float = 0.95, |
|
top_k: int = 40, |
|
repetition_penalty: float = 1.2, |
|
) -> Iterator[str]: |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
for char in "help": |
|
yield "help" |
|
|
|
|
|
chat_interface = ChatInterface( |
|
fn=generate, |
|
prefence_techniques="dpo", |
|
min_turns=1, |
|
max_turns=10, |
|
repo_id="llm-human-feedback-collector-chat-interface-dpo", |
|
chatbot=gr.Chatbot(height=450, label="Meta-Llama-3.1-8B-Instruct", show_share_button=True), |
|
cache_examples=False, |
|
additional_inputs=[ |
|
gr.Slider( |
|
label="Max new tokens", |
|
minimum=1, |
|
maximum=MAX_MAX_NEW_TOKENS, |
|
step=1, |
|
value=DEFAULT_MAX_NEW_TOKENS, |
|
), |
|
gr.Slider( |
|
label="Temperature", |
|
minimum=0.05, |
|
maximum=1.2, |
|
step=0.05, |
|
value=0.2, |
|
), |
|
gr.Slider( |
|
label="Top-p (nucleus sampling)", |
|
minimum=0.05, |
|
maximum=1.0, |
|
step=0.05, |
|
value=0.9, |
|
), |
|
gr.Slider( |
|
label="Top-k", |
|
minimum=1, |
|
maximum=1000, |
|
step=1, |
|
value=50, |
|
), |
|
gr.Slider( |
|
label="Repetition penalty", |
|
minimum=1.0, |
|
maximum=2.0, |
|
step=0.05, |
|
value=1.2, |
|
), |
|
], |
|
examples=[ |
|
["""What word doesn't make sense in this row: "car, airplane, lama, bus"?"""], |
|
["Write a news article about the usage of Lama's by the CSI"], |
|
["What are great things cook when getting started with Asian cooking?"], |
|
["Who was Anthony Bourdain?"], |
|
], |
|
title="💪🏽🦾 LLM human-feedback collector ChatInterface (DPO) 🦾💪🏽", |
|
description="""This is an adaptation of the gr.ChatInferface which allows for human feedback collection for SFT, DPO and KTO.""", |
|
) |
|
|
|
with gr.Blocks(css="style.css") as demo: |
|
chat_interface.render() |
|
|
|
if __name__ == "__main__": |
|
demo.queue(max_size=20).launch() |
|
|