mlboydaisuke's picture
ColModernVBERT -> Core AI: query + doc encoders (fp16/fp32), visual document retrieval (late-interaction / MaxSim)
2175916 verified
Raw
History Blame Contribute Delete
479 Bytes
{
"description" : "ColModernVBERT visual document retriever (document\/image encoder, single 512px tile): SigLIP2 vision + pixel-shuffle x4 -> ModernBERT-150M -> custom_text_proj(768->128) -> per-token L2-normalized 128-d multi-vector for ColBERT-style late-interaction (MaxSim). https:\/\/huggingface.co\/ModernVBERT\/colmodernvbert",
"creationDate" : "20260624T025647Z",
"license" : "MIT",
"assetVersion" : "2.0",
"author" : "ModernVBERT (Illuin Technology \/ ETH)"
}