Nvidia Releases NVLM-1.0-D-72B Multimodal LLM with Decoder-Only Architecture Achieving SOTA Results on Vision-Language Tasks

Oct 1, 2024, 05:58 AM

Nvidia has released NVLM-1.0-D-72B, a frontier-class multimodal large language model (LLM) with a decoder-only architecture. The model achieves state-of-the-art (SOTA) results on both vision-language and text-only tasks. It is reported to rival other advanced models such as GPT-4o, Llama 3-V 405B, and InternVL 2 in various evaluations, including math and coding. Nvidia has also made the checkpoint and inference scripts available on Hugging Face, with training code and additional versions like NVLM-1.0-X and NVLM-1.0-H expected to follow.