SupraLabs has released Supra2.5-65M, a new family of small language models, including base and instruction-tuned versions. The 65 million parameter model is designed to be highly efficient, capable of running on minimal hardware, and is reportedly better than the lab's previous 50M and 100M models. The training was conducted locally over four days using two RTX A6000 GPUs on a 100 billion token dataset, with benchmarks available in an attached image. AI
IMPACT Provides a new, highly efficient small model option for users with limited hardware resources.
RANK_REASON Release of a new small language model by a non-frontier lab. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →