I also seperately provide just the prompts in prompts.json keys are the image id, and the values are the captions generated Captions generated by moondream: vikhyatk/moondream2 Latents generated by SDXL VAE: madebyollin/sdxl vae fp16 fix Embeddings generated by SigLIP: hf hub:timm/ViT SO400M 14 SigLIP 384 Original dataset: common canvas/commoncatalog cc by Latents f32 and embeddings are f16 bytes Compute cost: 16x3090 for 3 day. Approximately.
We use cookies for essential functionality and analytics. You can accept or reject analytics cookies.Cookie policy