The field of large language models continues to swell as Stability AI, maker of the popular image generation tool Stable Diffusion, has launched a suite of open-source language model tools. The publicly available alpha version of the suite, called StableLM, currently contains models with 3 billion and 7 billion parameters, with 15 billion, 30 billion, and 65 billion parameter models marked as "work in progress," and 175-billion parameters for future development. billion-parameter model.
In comparison, OpenAI's GPT-4 has an estimated parameter count of 1 trillion, 6x more than its predecessor, GPT-3. However, as Stability AI points out in its blog post announcing StableLM, parameter counts may not be a good metric for measuring the power of large language models (LLMs):
“StableLM is trained on a new experimental dataset built on The Pile, but the dataset is three times larger at 1.5 trillion content tokens. The richness of this dataset enables StableLM to perform better on conversational and encoding tasks. shows astonishingly high performance in , despite its small number of parameters, only 3 to 7 billion.” It is unclear exactly how robust the StableLM model is. The Stability AI team notes on the organization's GitHub page that more information on LLM's capabilities, including model specifications and training settings, will be forthcoming.
If these models perform well enough in tests, the emergence of a robust open-source alternative to OpenAI’s ChatGPT could interest the cryptocurrency trading community. People are building advanced trading bots on top of the GPT API and new variants that include access to third-party tools such as BabyAGI and AutoGPT.
For tech-savvy traders who don't want to pay for access to OpenAI, adding an open-source model to the mix could be a boon. Those interested can test the live interface to the 7 billion parameter StableLM model hosted on HuggingFace. However, as of the time of this post, attempts to do so have found the site to be overwhelmed or at full capacity.
