H Company Releases NeoMME: A Family of 260M and 800M Single-Tower Multimodal Encoders That Drop the Vision Tower and Causal Decoder
MarkTechPost Asif Razzaq
H Company released NeoMME, a 260M/800M single-tower multimodal encoder family that removes both the separate vision tower and the causal decoder used in many production-style retrievers. The 260M NeoMME-Retriever reaches 0.523 nDCG@10 on ViDoRe v3, while the model indexes 51.3 pages per second on a single NVIDIA L40S and the checkpoint is available under Apache 2.0 on Hugging Face. As a result, retrieval indexing storage is cut from about 1.5 MB to 39.0 kB per page (with factor-10 pooling and int8) and the models can be deployed without day-one integration work in Transformers.
Why it matters
We look at NeoMME, a family of 260M and 800M bidirectional encoders from H Company. Unlike ColPali-style retrievers, it processes multilingual text tokens and raw 32×32 image patches in a single Transformer, with no pretrained vision tower and no causal decoder. We cover the masked discrete-diffusion pretraining objective, the dual dense and late-interaction retrieval heads, and the ViDoRe v3 results where the 260M model reaches 0.523 nDCG@10. We also break down the 255× index compression, the 51.3 pages per second indexing throughput on one L40S, and the text-retrieval gaps the authors acknowledge. The post H Company Releases NeoMME: A Family of 260M and 800M Single-Tower Multimodal Encoders That Drop the Vision Tower and Causal Decoder appeared first on MarkTechPost.
Related stories
Liquid AI Releases LFM2.5-Encoder-230M and LFM2.5-Encoder-350M: Bidirectional Encoders That Stay Fast at 8K Context on CPU
MarkTechPost · 1 month ago ·
32