metadata
license: apache-2.0
language:
- en
These are the weights for OLIVE trained on the refCOCOg dataset. The LLM decoder used is the huggingface instruction-tuned llama 2 checkpoint, and the vision encoder is a CLIP Vision Transformer with 14x14 patch size and 336 pixel resolution. For more details on training and usage, check out the github repository at https://github.com/tossowski/Olive.