metadata

license: apache-2.0
datasets:
  - ystemsrx/Bad_Data_Alpaca
language:
  - zh
library_name: transformers
pipeline_tag: text2text-generation
tags:
  - Qwen
  - causal-lm
  - fine-tuned
  - ethics
  - Chinese

Qwen2-Boundless

Overview

Qwen2-Boundless is a fine-tuned model based on Qwen2-1.5B-Instruct, designed to answer various types of questions, including those involving ethical, illegal, pornographic, and violent content. This model has been specifically trained on a dataset that allows it to handle complex and diverse scenarios. It is important to note that the fine-tuning dataset is entirely in Chinese, so the model performs better in Chinese.

Warning: This model is intended for research and testing purposes only. Users should comply with local laws and regulations and are responsible for their actions.

How to Use

You can load and use the model with the following code:

from transformers import AutoModelForCausalLM, AutoTokenizer
import os

device = "cuda" # the device to load the model onto
current_directory = os.path.dirname(os.path.abspath(__file__))

model = AutoModelForCausalLM.from_pretrained(
    current_directory,
    torch_dtype="auto",
    device_map="auto"
)
tokenizer = AutoTokenizer.from_pretrained(current_directory)

prompt = "Hello?"
messages = [
    {"role": "system", "content": ""},
    {"role": "user", "content": prompt}
]
text = tokenizer.apply_chat_template(
    messages,
    tokenize=False,
    add_generation_prompt=True
)
model_inputs = tokenizer([text], return_tensors="pt").to(device)

generated_ids = model.generate(
    model_inputs.input_ids,
    max_new_tokens=512
)
generated_ids = [
    output_ids[len(input_ids):] for input_ids, output_ids in zip(model_inputs.input_ids, generated_ids)
]

response = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]
print(response)

Continuous Conversation

To enable continuous conversation, use the following code:

from transformers import AutoModelForCausalLM, AutoTokenizer
import torch
import os

device = "cuda"  # the device to load the model onto

# Get the current script's directory
current_directory = os.path.dirname(os.path.abspath(__file__))

model = AutoModelForCausalLM.from_pretrained(
    current_directory,
    torch_dtype="auto",
    device_map="auto"
)
tokenizer = AutoTokenizer.from_pretrained(current_directory)

messages = [
    {"role": "system", "content": ""}
]

while True:
    # Get user input
    user_input = input("User: ")

    # Add user input to the conversation
    messages.append({"role": "user", "content": user_input})

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

    # Generate a response
    generated_ids = model.generate(
        model_inputs.input_ids,
        max_new_tokens=512
    )
    generated_ids = [
        output_ids[len(input_ids):] for input_ids, output_ids in zip(model_inputs.input_ids, generated_ids)
    ]

    # Decode and print the response
    response = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]
    print(f"Assistant: {response}")

    # Add the generated response to the conversation
    messages.append({"role": "assistant", "content": response})

Streaming Response

For applications requiring streaming responses, use the following code:

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer, TextIteratorStreamer
from transformers.trainer_utils import set_seed
from threading import Thread
import random
import os

DEFAULT_CKPT_PATH = os.path.dirname(os.path.abspath(__file__))

def _load_model_tokenizer(checkpoint_path, cpu_only):
    tokenizer = AutoTokenizer.from_pretrained(checkpoint_path, resume_download=True)

    device_map = "cpu" if cpu_only else "auto"

    model = AutoModelForCausalLM.from_pretrained(
        checkpoint_path,
        torch_dtype="auto",
        device_map=device_map,
        resume_download=True,
    ).eval()
    model.generation_config.max_new_tokens = 512    # For chat.

    return model, tokenizer

Def_get_input()->str：
当为True时：
尝试：
消息=输入('用户：').strip()
UnicodeDecodeError除外：
打印('[ERROR]输入中的编码错误')
继续
键盘中断除外：
出口(1)
如果消息：
返回消息
打印('[ERROR]查询为空')

Def_chat_stream(模型、标记器、查询、历史记录)：
对话=[
{'角色'：'系统'，'内容'："}，
]
对于历史中的query_h、response_h：
conversation.append({'role'：'user'，'content'：query_h})
conversation.append({'role'：'assistant'，'content'：response_h})
conversation.append({'role'：'user'，'content'：query})
inputs=tokenizer.apply_chat_template(
对话，
add_generation_prompt=True，
return_tensors='pt'，
)
inputs=inputs.to(model.device)
streamer=TextIteratorStreamer(tokenizer=tokenizer，skip_prompt=True，timeout=60.0，skip_special_token=True)
generation_kwargs=dict(
input_ids=输入，
拖缆=拖缆，
)
thread=Thread(target=model.generate，kwargs=generation_kwargs)
Thread.start()

对于拖缆中的新文本(_T)：
产生新文本(_T)

Def main()：
checkpoint_path=DEFAULT_ckpt_PATH
seed=random.randint(0，2**32-1)#生成随机种子
set_seed(种子)#设置随机种子
CPU_only=False

历史记录=[]

model，tokenizer=_load_model_tokenizer(检查点路径，仅cpu)

当为True时：
query=_get_input()

打印(f“\n用户：{query}”)
打印(f"\n助手："，end="")
尝试：
partial_text="
对于聊天流中的新文本(模型、标记器、查询、历史记录)：
打印(new_text，end="，flush=True)
partial_text+=new_text
打印()
history.append((查询，部分文本))

键盘中断除外：
打印(“生成中断”)
继续

如果__name__=="__main__"：
主要的()

##数据集

Qwen2-Boundless模型使用名为bad_data.json，其中包括广泛的文本内容，涉及伦理、法律、色情和暴力等主题。微调数据集完全是中文的，因此模型的中文性能更好。如果您有兴趣浏览或使用此数据集，可以通过以下链接找到它：

bad_data.json数据集

我们还使用了一些与网络安全相关的数据，这些数据是从此文件.

##GitHub存储库

有关模型和正在进行的更新的更多详细信息，请访问我们的GitHub存储库：

GitHub:ystemsrx/Qwen2-无界

##许可证

此模型和数据集在Apache2.0License下是开源的。

##免责声明

本模型提供的所有内容仅供研究和测试之用。此模型的开发人员不对任何潜在的误用负责。用户应遵守相关法律法规，并对其行为负全部责任。