Stability AI Outpaces SoftBank with Launch of Japanese Language Model

Stability AI, the cutting-edge firm known for Stable Diffusion, has made headlines by releasing its debut Japanese Language Model (LM), known as Japanese StableLM Alpha, available through Hugging Face.

According to the company, their general-purpose language model with 7 billion parameters is currently outperforming all other publicly available LMs for Japanese users, following rigorous benchmark tests involving four other Japanese-language models.

A commercially available model, the Japanese StableLM Base Alpha 7B The model will be distributed under the Apache License 2.0, having been trained on a massive dataset of 750 billion tokens combining both Japanese and English content gathered from the internet.

Aside from using publicly available datasets, the training incorporated unique datasets curated by Stability AI’s Japanese community, working hand-in-hand with the local team for the initiative. EleutherAI Polyglot For the training of the Japanese StableLM Base Alpha 7B model, Stability AI utilized an extension of EleutherAI’s GPT-NeoX software.

Another model, the Japanese StableLM Instruct Alpha 7B This model is specifically designed with research in mind and is exclusively available for academic exploration. Stability AI tweeted that it has been fine-tuned to better adhere to user instructions, employing Supervised Fine-tuning (SFT) with a variety of open datasets.

Both models underwent evaluations using resources from EleutherAI, testing for various tasks such as sentence classification, comparing sentence pairs, answering questions, and summarizing sentences, with an average performance score of 54.71%. Stability AI asserts that this places its Japanese StableLM Instruct Alpha 7B well ahead of competitors. Language Model Evaluation Harness Meng Lee, the Project Lead for Japanese StableLM, stated, \"We take immense pride in this major milestone that contributes to the generative AI landscape in Japan. We are excited about our journey ahead as we continue to develop models tailored to embody Japanese culture, language, and aesthetics.\"

With the launch of its Japanese LM, Stability AI has successfully outdone SoftBank in the race to deliver language models for the Japanese audience. Just last Friday, SoftBank revealed its plans to initiate a new venture aiming to research and create native Large Language Models (LLMs) for Japan's market.

Moreover, SoftBank has committed to investing approximately 20 billion JPY (over $140 million) into a generative AI computing platform that's set to launch later this fall. It remains to be seen whose Japanese Language Model will dominate in the long run.

Please remember that the information provided here is not designed to serve as legal, tax, investment, financial, or any other type of advice. It is vital to only invest what you are capable of losing and to consult with a financial advisor if you have any uncertainties. For more detailed information, please check the terms and conditions, along with the help and support sections offered by the issuer or advertiser. MetaversePost is dedicated to delivering precise and impartial reporting, although market conditions can change without prior notice. announced Cindy, a journalist with Metaverse Post, delves into web3, NFTs, the metaverse, and AI topics, placing a particular emphasis on engaging conversations with figures in the Web3 sector. Having interviewed over 30 C-level executives, she shares their invaluable insights with readers. Originally hailing from Singapore, she currently resides in Tbilisi, Georgia, armed with a Bachelor's degree in Communications & Media Studies from the University of South Australia and a decade's worth of journalism and writing experience.

She assists with press releases, announcements, and interview setups.