Lion Rock Laboratory open-sources Yuwu embodied reasoning model

The Lion Rock Artificial Intelligence Laboratory under the China Merchants Group Advanced Technology Research Institute released and open-sourced the embodied reasoning vision-language model Yuwu Lion-ER-35B-A3B. The model uses a MoE architecture with approximately 35B total parameters and about 3B activated per token; its weights have been made available on Hugging Face in BF16 Safetensors format and are natively compatible with inference frameworks such as Transformers, vLLM, and SGLang. The laboratory states that on ten public embodied-related benchmarks, the model improves on the base model across all metrics, with an equally weighted composite average score of 65.29. This result is an evaluation provided by the publisher.

AI 根据报道生成 · 7 days ago更新