recursal/Finch-MoE-37B-A11B-v0.1-HF

The recursal/Finch-MoE-37B-A11B-v0.1-HF is a Mixture of Experts (MoE) model developed by Recursal, based on the RWKV architecture. This 37 billion parameter model (with 11 billion active parameters) is designed for general language tasks, demonstrating improved performance over previous Finch and Eagle models on benchmarks like ARC-C, MMLU, and Winogrande. Its MoE architecture aims to provide enhanced capabilities while potentially optimizing inference efficiency compared to dense models of similar overall parameter count.