New LLM by StabilityAI | | | | Turtles AI

  #StabilityAI, the leading #AI research company behind #StableDiffusion for image creation, has announced the release of its #StableLM series of large language models (#LLM), with the initial set of StableLM-alpha models now available for use. The new models are built on The Pile, a massive #dataset containing 1.5 trillion tokens, which is three times larger than the previous dataset used by StabilityAI. The StableLM-alpha models are currently available in two sizes: 3B and 7B parameters. There are also plans to release 15B and 30B models in the near future. The base models are released under CC BY-SA-4.0, making them freely available for use. In addition to the StableLM-alpha models, StabilityAI has also fine-tuned the model with Stanford Alpaca's procedure using a combination of five recent datasets for conversational agents, including Stanford's Alpaca, Nomic-AI's gpt4all, RyokoAI's ShareGPT52K datasets, Databricks labs' Dolly, and Anthropic's HH. These models will be released as StableLM-Tuned-Alpha. Users can try out the 7B StableLM-Tuned-Alpha model right now on Hugging Face Spaces. The models will be continuously updated with new checkpoints, and an upcoming technical report will document the model specifications and the training settings. All StableLM models are hosted on the Hugging Face hub. Users can check out a notebook to run inference with limited GPU capabilities. To get started chatting with StableLM-Tuned-Alpha, users can use the provided code snippet. While the StableLM models offer exciting possibilities, there are also some potential issues to consider. As with any large language model, without additional fine-tuning and reinforcement learning, the responses a user gets might be of varying quality and could potentially include offensive language and views. However, StabilityAI is confident that with better data, community feedback, and optimization, these issues can be minimized. StabilityAI is excited about the potential of its StableLM series to help users with a wide range of tasks, from writing poetry and short stories to writing code. The company is also open to ideas for future development and encourages users to reach out on their Discord channel with suggestions.