Ecosystem
A 73.82 MB EmbeddingGemma 2 checkpoint squeezes multimodal embeddings below one bit per weight
Community reports 4 sources
E
EmbeddingGemma 2 0.8 BPW Multimodal
Model Oct 15, 2026
lethalbeats' community EmbeddingGemma 2 quantization with a 73.82 MB weight file, a custom CPU PyTorch loader and author-reported multimodal inference.
Visit primary link
2 sources attached
What happened lately
EmbeddingGemma 2 0.8 BPW Multimodal timeline
This community EmbeddingGemma 2 variant uses binary latent-factor quantization inspired by LittleBit, selected higher-precision weights and a rank-4 text-query adapter. Its manifest lists 744,371,992 parameters and 0.79332 bits per weight. The 73.82 MB figure measures the stored weight checkpoint, not total download size or runtime memory.
The included PyTorch loader accepts text and, according to the author, images, video and audio on CPU. The card reports reduced-corpus FiQA text retrieval with lower Recall@10 than the base model. It does not measure multimodal retrieval quality, quality outside FiQA or quality at the optional truncated embedding dimensions. Repository creation on October 9 predates the current October 15 checkpoint; an update timestamp alone does not establish its first release date.