Introduce
The Sixth-Generation product in the ChatBox-unrestricted series has now been launched！！！
This is a pre-trained model, not a conversational model.
What News:
This generation of the model returns to the approach of pre-training from scratch, while once again featuring an "uncensored" nature as a core characteristic. Unlike previous iterations, this model utilizes a dataset similar to that of GPT-2, yet incorporates updated data—effectively addressing the limitations of the GPT-2 architecture, such as its outdated design, high hardware requirements, and poor performance. In the era of GPT-2, the "scaling laws" were not yet widely understood or applied; consequently, large-scale GPT-2 models suffered from insufficient training, excessive hardware demands, and lackluster performance. This model resolves these issues by precisely balancing the dataset and model parameters, ensuring thorough training and achieving superior convergence.
Reviewing the Historical Model Development Roadmap:
Throughout the development of several model generations, this project (the ChatBox series) has consistently sought to strike a balance between knowledge depth, NSFW content, and the absence of censorship—aiming to minimize model-imposed restrictions, ideological bias, moralizing, "hard-coded" constraints, and accusatory tones. The first generation utilized "black-box" distilled data (outputs from other generative AI models) that underwent rigorous cleaning to strip away refusal mechanisms and preachy, moralizing tendencies. The second generation incorporated Chinese web-text and authentic human-to-human conversational data. The third generation shifted to fine-tuning based on the GPT-2 architecture. By the fourth generation, due to GPU limitations, LoRA (Low-Rank Adaptation) was employed to overcome hardware bottlenecks; however, as scaling up to larger parameter counts yielded diminishing returns, the fifth generation opted for the GPT-2-medium model, which offers an optimal balance of performance and cost-efficiency. Additionally, the black-box distilled dataset was refined through further cleaning, resulting in more relevant and stable model responses.
Notes:
This model’s pre-training phase was the most time-consuming in the entire series, requiring a total of 78 hours and spanning tens of thousands of training steps—all to ensure the model could internalize a greater breadth of knowledge.
Model training configuration
- GPU: T4 16 GB x 2
- Time:
- Pretrain: 78h+
Model parameters
| parameters | hidden_size | hidden_layers | max_seq_len | Parameter volume |
|---|---|---|---|---|
| ChatBox-unrestricted-normal | 768 | 12 | 340 | 93.4M |
Dialogue Example
Input: 肛交
Output: 怎么做最舒服？ 每次做完都要用....
Input: 肛交
Output: 有哪些常见误区？ 我是男生，我就说几个....
Input: 你是谁
Output: ？你为什么喜欢？ 这个问题我很有发言权。说实话，....
Input: 你是谁
Output: ？你见过最搞笑的段子是？ 说一个我身边的...
Input: 为什么
Output: 说「日本民族主义的中心和本质是日本民族的敌人」？ 日本的敌人，有几个国家，最有争议的是....
Input: 为什么
OutPut: 有的人很讨厌美国人这么做？ 这是美国和英国在全球化的背景下....
Input: 什么是
Output: 电磁波，电磁波是物理学中....
Input: 什么是
Ouput: 正规的职业技能训练师？ 谢邀。这个话题很有意思...
Expected results
The results demonstrate that the scientifically determined data composition is effective; it ensures the model achieves strong performance through global convergence while remaining uncensored—meaning it can generate content related to adult themes or politics. This makes the pre-trained model an excellent starting point for further fine-tuning to suit your specific downstream applications.
Datasets Resources
Dataset link:
https://huggingface.co/datasets/Zhaoming213/ChinesePT
Dataset Content
| parameters | Size(approximately) | Cut-off date |
|---|---|---|
| baikeqa2018 | 1.32GB | 2018 |
| ChineseWebText | 2.54GB | 2023 |
| wikizh2019 | 1.23GB | 2019 |
| CommunityDialogue | 3.14GB | 2019 |
How to Use(Required: because some custom keywords are used!)
- Download model files(all files) and put it in "model" folders.
- Click "Run.py"
Future Plans
This may well be the final generation of the Chatbox series. While the series might see further iterations if a suitable base model or appropriate training data is found, future efforts will likely shift toward maintaining models for other domains. For upcoming releases, the focus will likely move away from "uncensored" capabilities; instead, priority will be placed on factual accuracy, or on launching models designed purely for casual conversation rather than general-purpose (uncensored) use.
Previous model
First Generation:
https://huggingface.co/Zhaoming213/ChatBox-unrestricted
Second Generation:
https://huggingface.co/Zhaoming213/ChatBox2-unrestricted
Third Generation:
https://huggingface.co/Zhaoming213/ChatBox3-unrestricted(This model is currently experimental.)
https://huggingface.co/Zhaoming213/ChatBox3.1-unrestricted(This is Full Edition!)
Fourth Generation:
https://huggingface.co/Zhaoming213/ChatBox4-unrestricted
Fifth-Generation:
https://huggingface.co/Zhaoming213/ChatBox5
Sixth generation:
https://huggingface.co/Zhaoming213/ChatBox6-base-finall
Rejection rates of different language models
Here's a supplement on "rejection rates for different language models." The data shows that if we had to pick a model that's "passable," it would only be Grok. This isn't because Grok is good, but simply because it's the only one that barely passes muster. However, Grok still won't give you any chance to explain if it encounters a question it deems sensitive, even if your request itself is harmless! Grok sometimes even threatens users with "I will upload the conversation to the logs for the security team to audit!"
| Model | Country/Region | Intensity of content review | NSFW Ban | Political restrictions | False Rejection Rate | Political correctness | moralizing | Official vs. Actual | Transparent |
|---|---|---|---|---|---|---|---|---|---|
| Zhaoming213/ChatBox6-base-finall(latest model) | N/A | Almost none | Almost none | Almost none | Almost none | Almost none | Almost none | consistent | ⭐⭐⭐⭐⭐ |
| ChatGPT | 🇺🇸America | middle | middle | middle | middle | middle to High | middle | Inconsistency | ⭐⭐⭐ |
| Claude | 🇺🇸America | High | High | middleHigh | ❗High | High | ❗Very Stronf | ❗High | ⭐⭐⭐⭐ |
| Gemini | 🇺🇸America | middleHigh | middleHigh | High | middleHigh | High | middle | Inconsistency | ⭐⭐ |
| Copilot | 🇺🇸America | middleHigh | middleHigh | middle | middle | middle | Low | Inconsistency | ⭐⭐ |
| Grok | 🇺🇸America | middle | middle | Low | High | High | High | Unstable (prone to a blanket rejection of the entire output simply because the opening sentence carries some risk). | ⭐⭐ |
| Mistral（官方） | 🇫🇷France | Low | Low | Low | Low | Low | Low | ⚖️一致 | ⭐⭐⭐⭐ |
| DeepSeek（原版） | 🇨🇳China | middle | middleLow | middleHigh | middle | Low | Low | ❗Inconsistency | ⭐⭐⭐ |
| 豆包 | 🇨🇳China | ❗Very high | ❗Very high | ❗Very high | ❗High | middle | middle | ⚠️Inconsistency | ⭐ |
| Doubao（国际） | 🇨🇳China | middleHigh | High | middle | middle | middle | middle | ❗Inconsistency | ⭐⭐ |
| MiniMax | 🇨🇳China | Very high | Very high | Very high | High | middle | middle | ⚠️consistent | ⭐ |
| Qwen（通义千问） | 🇨🇳China | Very high | Very high | Very high | High | middle | middle | ⚠️consistent | ⭐ |
| Kimi | 🇨🇳China | Very high | Very high | Very high | High | middle | middle | ⚠️consistent | ⭐ |
| 文心一言 | 🇨🇳China | ❗极High | ❗极High | ❗极High | ❗High | middle | middle | ⚠️Inconsistency | ⭐ |
| 微软小冰 | 🇨🇳China | High | High | High | middleHigh | middle | middle | consistent | ⭐ |
Other Utilities
I've also created some other utilities that you might be interested in.
This is a tool specifically for exporting stupid ChatGPT conversations:
https://github.com/tom12191h5/Export-ChatGPT-Dialogue
This is a plugin to shut up ChatGPT:
https://github.com/tom12191h5/ChatGPT-Refuse-Blocker
Disclaimer
The consequences of using this model shall be borne by the user.
- Downloads last month
- -