Once ChatGPT was released, it caused a sensation in the AI circle. GPT-4 and chagpt-plugin opened a wider gap between OpenAI and other peers. For some reason, we may not be able to experience the magic of ChatGPT for ourselves. But now, if you haven't experienced ChatGPT, it's like not buying Bitcoin 10 years ago (the price has increased tens of thousands of times). So, I found a few ChatGPT replacements that will allow you to ride the current wave of AI.
First of all, it is stated that the following replacement versions need to be downloaded, installed and compiled by themselves, and certain programming skills are required, but the advantage is that once you start it, you will have your own personal GPT assistant.
Generative Large Language Models
alpaca.cpp
The project, jointly launched by OpenAI and Stanford, aims to enable users to run generative large language models locally.
The following is an unaccelerated rendering:
There are 3 versions of this project to choose from according to the size of the computer's memory. After testing, even the 智障version with the lowest memory usage can do a good job in answering factual questions, and 大聪明the version with 30GB of memory can already provide you with professional consultation . Unfortunately, questions are currently only available in English.
call.cpp
This project is based on the C/C++ version ported from Facebook's open source large language model facebookresearch / llama . The model is also available in the following versions according to the size of the computer memory:
| model | original size | quantized size (4-bit) |
|---|---|---|
| 7b | 13 GB | 3.9 GB |
| 13b | 24 GB | 7.8 GB |
| 30B | 60 GB | 19.5 GB |
| 65b | 120 GB | 38.5 GB |
The following is the rendering of the operation:
Multimodal Generative Models
In addition to AI assistants that can generate text, image generation is also an AI model that has received much attention.
Text to image: stable-diffusion
stable-diffusion is an AI model for converting text to images, and Wenxin Yige released by Baidu is also developed based on this open source model. You can build your own from the following address 文心一格.
The project is built based on the stable-diffusion model, not only has a user interface, but also can install other effect plug-ins as needed.
After the installation is complete, not only the language-to-image conversion function can be realized, but also pictures can be uploaded, and new images can be generated according to the pictures and your description.
Speech to Text: Whisper
Whisper is a speech-to-text AI model open sourced by OpenAI.
After installing this model, you can have the ability to convert voice/video to text.
And if you are interested enough, you can also combine Whisper with the text generation model mentioned above, so that you have an intelligent robot that can talk.
The above has introduced a variety of open source AI models that can be built. Although building your own GPT requires some programming skills and time, it will be very useful to have a private and powerful language model assistant by yourself. value things. And this is also an interesting and useful opportunity to explore, to understand and experience ChatGPT in depth, and to build your own language model, so as to gain more development opportunities in the field of AI.