|
import gradio as gr |
|
import torch |
|
from transformers import AutoTokenizer, AutoModelForCausalLM |
|
import warnings |
|
warnings.filterwarnings("ignore") |
|
|
|
""" |
|
デバッグ用のシンプルなSarashinaチャットボット |
|
additional_inputsなしでテスト |
|
""" |
|
|
|
|
|
MODEL_NAME = "sbintuitions/sarashina2.2-3b-instruct-v0.1" |
|
|
|
print("モデルを読み込み中...") |
|
tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME, trust_remote_code=True) |
|
model = AutoModelForCausalLM.from_pretrained( |
|
MODEL_NAME, |
|
torch_dtype=torch.float16 if torch.cuda.is_available() else torch.float32, |
|
device_map="auto" if torch.cuda.is_available() else None, |
|
trust_remote_code=True |
|
) |
|
print("モデルの読み込みが完了しました。") |
|
|
|
def respond(message, history): |
|
""" |
|
シンプルなチャットボット応答関数 |
|
additional_inputsなし |
|
""" |
|
try: |
|
|
|
print(f"DEBUG - message: {message} (type: {type(message)})") |
|
print(f"DEBUG - history: {history} (type: {type(history)})") |
|
|
|
|
|
system_message = "あなたは親切で知識豊富な日本語アシスタントです。ユーザーの質問に丁寧に答えてください。" |
|
|
|
|
|
conversation = f"システム: {system_message}\n" |
|
|
|
|
|
if history and isinstance(history, list): |
|
for item in history: |
|
if isinstance(item, (list, tuple)) and len(item) >= 2: |
|
user_msg, bot_msg = item[0], item[1] |
|
if user_msg: |
|
conversation += f"ユーザー: {user_msg}\n" |
|
if bot_msg: |
|
conversation += f"アシスタント: {bot_msg}\n" |
|
|
|
|
|
conversation += f"ユーザー: {message}\nアシスタント: " |
|
|
|
|
|
inputs = tokenizer.encode(conversation, return_tensors="pt") |
|
|
|
|
|
if torch.cuda.is_available(): |
|
inputs = inputs.cuda() |
|
|
|
|
|
with torch.no_grad(): |
|
outputs = model.generate( |
|
inputs, |
|
max_new_tokens=512, |
|
temperature=0.7, |
|
top_p=0.95, |
|
do_sample=True, |
|
pad_token_id=tokenizer.eos_token_id, |
|
eos_token_id=tokenizer.eos_token_id, |
|
repetition_penalty=1.1 |
|
) |
|
|
|
|
|
generated = tokenizer.decode(outputs[0], skip_special_tokens=True) |
|
|
|
|
|
full_response = generated[len(conversation):].strip() |
|
|
|
|
|
if "ユーザー:" in full_response: |
|
full_response = full_response.split("ユーザー:")[0].strip() |
|
|
|
|
|
for i in range(len(full_response)): |
|
response = full_response[:i+1] |
|
yield response |
|
|
|
except Exception as e: |
|
yield f"エラーが発生しました: {str(e)}" |
|
|
|
""" |
|
シンプルなChatInterface(additional_inputsなし) |
|
""" |
|
demo = gr.ChatInterface( |
|
respond, |
|
title="🤖 Sarashina Chatbot (Simple)", |
|
description="Sarashina2.2-3b-instruct モデルを使用した日本語チャットボットです。(デバッグ用)", |
|
theme=gr.themes.Soft(), |
|
examples=[ |
|
"こんにちは!", |
|
"日本について教えて", |
|
"プログラミングの質問があります", |
|
], |
|
cache_examples=False, |
|
) |
|
|
|
if __name__ == "__main__": |
|
demo.launch( |
|
server_name="0.0.0.0", |
|
server_port=7860, |
|
share=False, |
|
show_api=True, |
|
debug=True |
|
) |