# pip install bitsandbytes

from transformers import AutoModelForCausalLM, AutoTokenizer, TextIteratorStreamer
import torch
import threading

from transformers import BitsAndBytesConfig
bnb_config = BitsAndBytesConfig(
    load_in_4bit=True,
    bnb_4bit_compute_dtype=torch.bfloat16,
    bnb_4bit_use_double_quant=True,
    bnb_4bit_quant_type="nf4"
)

# model_name = "netease-youdao/Confucius3-Math"
model_name = r"D:\Confucius3-Math\netease-youdao\Confucius3-Math"

SYSTEM_PROMPT_TEMPLATE = """用户与助手之间的对话。用户提出一个问题，助手予以解答。助手先在脑海中思考推理过程，然后为用户提供答案。推理过程和答案分别用 <think> </think> 和 <answer> </answer> 标签括起来，即 <think> 这里的推理过程 </think> <answer> 这里的答案 </answer> 。"""

USER_PROMPT_TEMPLATE = """{question}"""

question = "1+1=?"

model = AutoModelForCausalLM.from_pretrained(
    model_name,
    # torch_dtype="auto",
    quantization_config=bnb_config,
    device_map="auto"
)

tokenizer = AutoTokenizer.from_pretrained(model_name)
messages = [
    {'role': 'system', 'content': SYSTEM_PROMPT_TEMPLATE},
    {'role': 'user', 'content': USER_PROMPT_TEMPLATE.format(question=question)},
]

text = tokenizer.apply_chat_template(
    messages,
    tokenize=False,
    add_generation_prompt=True
)
model_inputs = tokenizer([text], return_tensors="pt").to(model.device)

# 创建流式输出器
streamer = TextIteratorStreamer(tokenizer, skip_special_tokens=True, timeout=10.0)

# 设置生成参数，添加streamer
generation_kwargs = {
    **model_inputs,
    "streamer": streamer,
    "max_new_tokens": 32768
}

# 创建线程来处理流式输出
thread = threading.Thread(target=model.generate, kwargs=generation_kwargs)
thread.start()

# 流式输出结果
print("流式输出开始：")
for chunk in streamer:
    if chunk:
        print(chunk, end="", flush=True)

thread.join()
print("\n流式输出结束")