TechFlow news: Web3 AI platform MyShell, in collaboration with MIT's Computer Science and Artificial Intelligence Laboratory, has released a new large language model (LLM) called JetMoE-8B. According to the research paper, its training cost is less than $100,000. The model outperforms Llama2-7B and even surpasses Llama2-13B-Chat on chat-based tasks—despite the latter having access to billions of dollars in training resources. JetMoE-8B is highly open and academic-friendly: it is trained solely on public datasets, its code is open-source, and requires no proprietary resources. It can be fine-tuned under limited computational budgets affordable to most labs, such as consumer-grade GPUs.
Add to Favorites
Share to Social Media
