3 ms·Blip-2: harvesting development of pretrained vision models for LLM training2 points by anigbrowl 4y ago