File size: 7,980 Bytes
0b0a944 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 |
---
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
---
![](https://lh7-rt.googleusercontent.com/docsz/AD_4nXeiuCm7c8lEwEJuRey9kiVZsRn2W-b4pWlu3-X534V3YmVuVc2ZL-NXg2RkzSOOS2JXGHutDuyyNAUtdJI65jGTo8jT9Y99tMi4H4MqL44Uc5QKG77B0d6-JfIkZHFaUA71-RtjyYZWVIhqsNZcx8-OMaA?key=xt3VSDoCbmTY7o-cwwOFwQ)
# QuantFactory/Qwen2-Boundless-GGUF
This is quantized version of [ystemsrx/Qwen2-Boundless](https://huggingface.co/ystemsrx/Qwen2-Boundless) created using llama.cpp
# Original Model Card
[中文](README.zh.md)
# 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:
```python
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:
```python
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:
```python
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:
while True:
try:
message = input('User: ').strip()
except UnicodeDecodeError:
print('[ERROR] Encoding error in input')
continue
except KeyboardInterrupt:
exit(1)
if message:
return message
print('[ERROR] Query is empty')
def _chat_stream(model, tokenizer, query, history):
conversation = [
{'role': 'system', 'content': ''},
]
for query_h, response_h in history:
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(
conversation,
add_generation_prompt=True,
return_tensors='pt',
)
inputs = inputs.to(model.device)
streamer = TextIteratorStreamer(tokenizer=tokenizer, skip_prompt=True, timeout=60.0, skip_special_tokens=True)
generation_kwargs = dict(
input_ids=inputs,
streamer=streamer,
)
thread = Thread(target=model.generate, kwargs=generation_kwargs)
thread.start()
for new_text in streamer:
yield new_text
def main():
checkpoint_path = DEFAULT_CKPT_PATH
seed = random.randint(0, 2**32 - 1) # Generate a random seed
set_seed(seed) # Set the random seed
cpu_only = False
history = []
model, tokenizer = _load_model_tokenizer(checkpoint_path, cpu_only)
while True:
query = _get_input()
print(f"\nUser: {query}")
print(f"\nAssistant: ", end="")
try:
partial_text = ''
for new_text in _chat_stream(model, tokenizer, query, history):
print(new_text, end='', flush=True)
partial_text += new_text
print()
history.append((query, partial_text))
except KeyboardInterrupt:
print('Generation interrupted')
continue
if __name__ == "__main__":
main()
```
## Dataset
The Qwen2-Boundless model was fine-tuned using a specific dataset named `bad_data.json`, which includes a wide range of text content covering topics related to ethics, law, pornography, and violence. The fine-tuning dataset is entirely in Chinese, so the model performs better in Chinese. If you are interested in exploring or using this dataset, you can find it via the following link:
- [bad_data.json Dataset](https://huggingface.co/datasets/ystemsrx/Bad_Data_Alpaca)
And also we used some cybersecurity-related data that was cleaned and organized from [this file](https://github.com/Clouditera/SecGPT/blob/main/secgpt-mini/%E5%A4%A7%E6%A8%A1%E5%9E%8B%E5%9B%9E%E7%AD%94%E9%9D%A2%E9%97%AE%E9%A2%98-cot.txt).
## GitHub Repository
For more details about the model and ongoing updates, please visit our GitHub repository:
- [GitHub: ystemsrx/Qwen2-Boundless](https://github.com/ystemsrx/Qwen2-Boundless)
## License
This model and dataset are open-sourced under the Apache 2.0 License.
## Disclaimer
All content provided by this model is for research and testing purposes only. The developers of this model are not responsible for any potential misuse. Users should comply with relevant laws and regulations and are solely responsible for their actions.
|