Sure can do! Here's passage reranking via submodular optimization with query "river and stream": diminishing return at 38 frames for the facility location function and at 6 frames for the saturated coverage function. Notebook to reproduce: colab.research.google.com/drive/1r8lO156…

jina-embeddings-v4-GGUF is here with different quantizations github.com/jina-ai/jina-e… Unsloth-like dynamic quants is on the way.

jina-reranker-m0-GGUF is here huggingface.co/jinaai/jina-re…

Your model doesn't dig 4K the way you do jina.ai/news/how-image… - For most visual document retrieval tasks, 1.2 megapixels resolution has the highest average score, and 2.4 megapixels is the best scoring resolution for the largest number of individual datasets. So this means: -