Stability AI continues to expand its list of AI services. The company already has generative neural networks for illustration in text queries and drafts (Stable Diffusion and Stable Doodle, respectively), but has now decided to compete with OpeAI and its ChatGPT.
At the moment this is just a trial version based on the current version of Stable Beluga. Anyone can try it, but you need to register or sign in to your Google account.
On Wednesday, Stability AI released a new family of open-source AI language models called StableLM. Stability hopes to replicate the catalytic effects of its open-source imaging model, Stable Diffusion, launched in 2022. Once perfected, StableLM could be used to create an open-source alternative to ChatGPT.
StableLM is currently available in alpha format on GitHub in model sizes of 3 billion and 7 billion metrics, where Stability says it can track model sizes of 15 billion and 65 billion metrics. The company releases models under the Creative Commons BY-SA-4 license.0, which requires customizations from the original creator and uses the same license.
Stability AI Ltd. is a London-based company that has become an open-source competitor to OpenAI, which despite its "open" name rarely releases open-source models and keeps its neural network weights confidential - a mass of numbers that define the base functionality of the AI model.
"Language models will be the foundation of our digital economy, and we want everyone to have a say in shaping them," Stability writes in an introductory blog post. “Models like StableLM demonstrate our commitment to transparent, accessible and useful AI technology."
Similar to GPT-4, the Large Language Model (LLM) supporting the most powerful version of ChatGPT, StableLM generates text by predicting the next token (word fragment) in a sequence. This sequence begins with information provided by the man in the form of "clues." This allows StableLM to compose human-like text and write programs.
Like other recent "small" LLMs like Metas LLaMA, Stanford Alpaca, Cerebras-GPT and Dolly 2.0, StableLM presumably achieves similar performance to the OpenAI GPT-3 reference model with far fewer metrics: 7 billion for StableLM versus 175 billion for GPT -3.
parameters are variables that the language model uses to learn from the training data. Fewer parameters make the language model smaller and more efficient, which can make it easier to run on local devices such as smartphones and laptops. However, achieving high performance with fewer parameters requires careful design, which is a major challenge in the field of artificial intelligence.
"Our StableLM models can generate text and code and support a range of downstream applications," says Stability. "They show how small, powerful models, with the right training, can become top performers."
According to Stability AI, StableLM was trained on a "new set of experimental data"; based on an open source dataset called The Pile, but three times larger. According to Stability, the "richness" of this data set, the details of which will be announced later, is responsible for the "surprisingly high performance". Patterns with smaller parameter sizes in conversational and programming tasks.
Last August, Stability funded and released the open-source launch of Stable Diffusion, developed by researchers in the CompVis group at the Ludwig Maximilian University of Munich.
As the first open-source latent diffusion model capable of generating images from prompts, Stable Diffusion ushered in an era of rapid development in image synthesis technology. It's also prompted major backlash from artists and companies, some of whom have sued Stability AI. Moving from stability to language models could lead to similar results.
users can test the 7 billion parameters of the StableLM Hugging Face base model and the refined model in Replicate. Additionally, Hugging Face supports a conversational friendly version of StableLM with a ChatGPT-like conversation format.
For More Technology Blog Visit Success Road