Public KG record
NeoMME-Retriever
Recorded overview
Dual-head design: dense head uses mean pooling, late-interaction head for fine-g
An evidence-backed entity from the project knowledge graph; detailed tool metadata is still being added.
Recorded details
- Benchmarks: ViDoRe v3 Pareto frontier for nDCG@10 and model size
- Benchmarks: 0.523 nDCG@10 (260M) and 0.556 (800M) on ViDoRe v3
- Benchmarks: 0.523 nDCG@10 on ViDoRe v3
- Benchmarks: 0.556 nDCG@10 on ViDoRe v3
- Benchmarks: ViDoRe v3
- Capabilities: returns dense and late-interaction embeddings in one forward pass
- Capabilities: ranks document page screenshots directly, bypassing OCR
- Capabilities: Retrieval-finetuned using mean-pooled dense head and late-interaction head
Evidence coverage
The KG currently records 8 known information dimensions and has 10 undocumented dimensions, linked to 2 public source articles. These counts describe record coverage, not model performance or verified fit.
Related public records
- ColQwen2.5-v0.2 — competes with
- NeoMME — based on
- ColQwen2.5-v0.2 — competes with
- Vultron Retriever Flash — competes with
- ViDoRe v3 — evaluated on