Nvidia's Nemotron 3.5 Super VL gets an open release in BF16 and NVFP4
ModelScope says its ms-swift toolkit supports inference and fine-tuning for the 120B-parameter multimodal model.
TLDR
ModelScope says Nvidia's Nemotron 3.5 Super VL is openly released in BF16 and NVFP4. It describes the model as a 120B-parameter mixture-of-experts system that activates 12B parameters per token and says ms-swift supports inference and fine-tuning. Separately, a post says its team partnered with Nvidia, Nolla and Nebius to post-train the model for dermatology diagnosis; the team says it set new standards on its internal measures of trainability.
Combined views
1.4K
1 Source, first seen ago