Beijing Academy of Artificial Intelligence and Tsinghua University unveiled a powerful AI model called EVA-CLIP-18B. It’s the largest of its kind, with 18 billion parameters, outperforming previous AI models by a large margin.
Features of EVA-CLIP-18B
EVA-CLIP-18B achieved an 80.7% zero-shot top-1 accuracy across 27 image classification benchmarks, surpassing its predecessors and other similar models. It was trained on a dataset of 2 billion image-text pairs, which is smaller than those used in other models. The approach to its training consistently improves performance without reaching saturation.
Benefits of EVA-CLIP-18B
EVA-CLIP-18B is versatile and effective, demonstrating its capabilities across multiple datasets, spanning image classification, image-text retrieval, and video classification tasks. EVA-CLIP-18B showcases outstanding performance, marking a significant advancement in CLIP model capabilities.
Brand New Model
EVA-CLIP-18B is the largest and highest-performing open-source CLIP model. It has proven to be powerful, versatile, and effective across various tasks. To check out the full paper of the research, visit this link and for more updates, follow their social media accounts.