11/12/2023 0 Comments Paper.io 4 online□ We released **LLaVA: Large Language and Vision Assistant**. Thanks to the community effort, LLaVA-13B with 4-bit quantization allows you to run on a GPU with as few as 12GB VRAM! Try it out ( ). □ We are releasing LLaVA-Lighting! Train a lite, multimodal GPT-4 with just $40 in 3 hours! See (#train-llava-lightning) for more details. We are releasing ( ), based on MPT-7B-Chat! See (#LLaVA-MPT-7b) for more details. We released **LLaVA-Med: Large Language and Vision Assistant for Biomedicine**, a step towards building biomedical domain large language and vision models with GPT-4 level capabilities. We released the preview for the most requested feature: DeepSpeed and LoRA support! Please see documentations (./docs/LoRA.md). ( ) on **Large Multimodal Models: Towards Building and Surpassing Multimodal GPT-4**! Please check out ( )] ( )] ( )] ( )]. We also support and verify training with RTX 3090 and RTX A6000. We release ( ) for benchmarking open-ended visual chat with results from Bard and Bing-Chat. □ We release a major upgrade, including support for LLaMA-2, LoRA training, 4-/8-bit inference, higher resolution (336x336), and a lot more. Further, if you are interested in the comprehensive review, evolution and trend of multimodal foundation models, please check out our recent survey paper ``Multimodal Foundation Models: From Specialists to General-Purpose Assistants''.
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |