Text | IQ dropped all over the place
In the early morning of the 20th, Stability AI released a new open source language model - StableLM, the company has developed a Stable Diffusion image generation tool. This news means that it is no longer limited to the field of image and video generation, and will officially join the text generation AI track.
StableLM models can generate text and code, and will power a range of downstream applications. The link below contains the StableLM series of language models being developed by Stability AI. Interested partners can try to chat with the model with 7 billion parameters in the link below Huggingface~
GitHub address :
https://github.com/stability-AI/stableLM/
Model official website :
https://stability.ai/
Huggingface Chat Address :
https://huggingface.co/spaces/stabilityai/stablelm-tuned-alpha-chat
StableLM
Stability AI has released StableLM on GitHub for developers to use as-is or after fine-tuning, and also released the RLHF-tuned model for research. StableLM follows the principle of open source and open to all, and returns the power to the developer community. Currently, StableLM is still in the Alpha stage, and only announced models with 3 billion and 7 billion parameters. The next step will be 15 billion to 65 billion parameters. larger parameter models.
According to reports, StableLM is trained on a new experimental data set based on The Pile, but the size of the data set is three times larger than the original, containing 1.5 trillion Token content.
Stability AI used the following dialogue to demonstrate the ability of their large model StableLM to generate text and code. It is really handy for English questions and answers. These questions and answers are generated by fine-tuning the model with 7 billion parameters:
In the dialogue experience with StableLM, the editor also found that it can communicate in Chinese and can basically understand my semantics (although the details are still problematic). It's a pity that it's still a bit far from ChatGPT. It knows a little Chinese, but not much , and starts talking nonsense when chatting...
Stability AI still emphasizes the desire for everyone to be able to participate in the design of its language model, and models like StableLM demonstrate a commitment to transparent, accessible and supportive AI technology:
Transparency : Open source models foster transparency, build trust, and help researchers work on explainable techniques to identify potential risks and help develop safeguards.
Ease of use : Models are easily accessible, and open and fine-grained access opportunities enable public research and academia to develop interpretable and secure techniques beyond what is possible with closed-source models.
Supportive : Designed to help users increase productivity, unleash creativity, and unlock new economic opportunities.
Finally, Stability AI reminds the potential problems of generated answers as follows:
Any pre-trained large-scale language model, without further fine-tuning and reinforcement learning, may have mixed responses, including offensive remarks and opinions. This will hopefully improve with scale, better data, community feedback and optimizations .
summary
Stability AI will release a full technical report shortly, and will also launch the RLHF program in the future and work with multiple communities to create open source datasets for AI Assistant. With the launch of the StableLM model suite, Stability AI is still continuing to promote the purpose of making basic AI technology available to everyone . Although it has been complained by many artists for alleged image infringement, it also received a round of financing at the end of last year and became A unicorn in the field of AI content generation.
There has always been a heated debate about whether the underlying tools of AI should be open source:
Some people think that open source will reduce the commercial value of the model, making it impossible for developers to get the rewards they deserve;
There are also those who believe that open source can promote the development and popularization of technology, and can better ensure its trust and transparency.
The open source of StableLM shows us the sincerity of Stability AI about " AI by the people, for the people ". Compared with some models that perform very well but whose black-box operation prevents ordinary users from knowing the details, the open source of StableLM not only helps to improve its transparency, but also gives more people the opportunity to develop and innovate on its basis .
With the continuous development of AI technology, more and more people are beginning to hope that these technologies can truly empower the public and unleash their creativity. Open source AI technology helps to achieve this goal, because it makes it easier for the technology to be improved and expanded, and it also enables more people to participate in the development and promotion of the technology . Further enable more people to benefit from it and create more value.
The author of the cute house: IQ dropped all over the place
I am studying for a master's degree in computer science at BIT. I am addicted to chatting with ChatGPT recently. I am curious about all novel NLP applications. I am trying to become a slash youth with a wide range of interests~
Recommended works
Reply keywords in the background [ join the group ]
Join the NLP, CV, search promotion and job hunting discussion group