DavidAU/Qwen3.6-27B-Fable-Fusion-711-Uncensored-Heretic-NM-DAU-NEO-MAX-MTP-GGUF
Image-Text-to-Text • 27B • Updated • 361k • 2.51k
Sachin Dohdiya
sachin7x
AI & ML interests
None yet
Recent Activity
liked a model 15 days ago
DavidAU/Qwen3.6-27B-Fable-Fusion-711-Uncensored-Heretic-NM-DAU-NEO-MAX-MTP-GGUF liked a model 23 days ago
Ngixdev/OmniCoder-Qwen3.5-9B-Claude-4.6-Opus-Uncensored-v2-GGUF liked a model 29 days ago
nerkyor/Qwen3.8-27B-Coder390-EfficientThink-Opus5.5-GPT6Astra-Grok4.7-DSV4Pro-K3-SFT-RLOO-MTP-DFlash2Organizations
None yet
liked a
model 15 days ago
liked a
model 23 days ago
liked a
model 29 days ago
liked a
dataset about 1 month ago
liked a
model about 1 month ago
reacted to Ryenhails's post with 🔥 about 1 month ago
Post
3,747
🚀 NanoVDR goes multi-vector: meet ColNanoVDR!
Multi-vector VLM retrievers lead visual document retrieval, but every search runs a multi-billion-parameter query encoder. We distill that encoder into a 149M text-only student that queries the teacher's existing page index directly. No re-indexing, and no pages during training.
🧠 How: OTW (Optimal Transport with Learned Weights) aligns the student's query tokens with the teacher's, even though the two tokenize differently (e.g. 17 vs 29 tokens). We prove the alignment cost bounds the MaxSim score gap on every page, so training only needs cached teacher query tokens.
📊 Five teachers → five 149M students, ViDoRe v3 NDCG@5:
- ColVec1.1-8b: 62.6 → 60.1 (96.0%)
- ColVec1.1-4b: 61.6 → 59.1 (95.8%)
- Vultron-4.5B: 61.0 → 58.3 (95.5%)
- ColQwen3.5-4.5B: 58.7 → 55.1 (93.8%)
- Tomoro-ColQwen3-8B: 59.0 → 54.9 (93.0%)
⚡ 26× faster query encoding on a single CPU thread (87 ms vs 2.3 s)
💾 Matches score distillation while reading 12.6× less cached teacher data
📄 Paper: ColNanoVDR: Document-Free Query Distillation for Multi-Vector Visual Document Retrieval via Optimal Transport (2609.34899)
🤗 Checkpoints (all five students):
nanovdr
💻 Code: https://github.com/Ryenhails/NanoVDR
🧩 Single-vector predecessor, NanoVDR: https://arxiv.org/abs/2603.12824
If you already serve one of these teachers, swap in the matching student and keep your index as is. Feedback and upvotes welcome! 🙌
Multi-vector VLM retrievers lead visual document retrieval, but every search runs a multi-billion-parameter query encoder. We distill that encoder into a 149M text-only student that queries the teacher's existing page index directly. No re-indexing, and no pages during training.
🧠 How: OTW (Optimal Transport with Learned Weights) aligns the student's query tokens with the teacher's, even though the two tokenize differently (e.g. 17 vs 29 tokens). We prove the alignment cost bounds the MaxSim score gap on every page, so training only needs cached teacher query tokens.
📊 Five teachers → five 149M students, ViDoRe v3 NDCG@5:
- ColVec1.1-8b: 62.6 → 60.1 (96.0%)
- ColVec1.1-4b: 61.6 → 59.1 (95.8%)
- Vultron-4.5B: 61.0 → 58.3 (95.5%)
- ColQwen3.5-4.5B: 58.7 → 55.1 (93.8%)
- Tomoro-ColQwen3-8B: 59.0 → 54.9 (93.0%)
⚡ 26× faster query encoding on a single CPU thread (87 ms vs 2.3 s)
💾 Matches score distillation while reading 12.6× less cached teacher data
📄 Paper: ColNanoVDR: Document-Free Query Distillation for Multi-Vector Visual Document Retrieval via Optimal Transport (2609.34899)
🤗 Checkpoints (all five students):
💻 Code: https://github.com/Ryenhails/NanoVDR
🧩 Single-vector predecessor, NanoVDR: https://arxiv.org/abs/2603.12824
If you already serve one of these teachers, swap in the matching student and keep your index as is. Feedback and upvotes welcome! 🙌
updated a
model 9 months ago
published a
model 9 months ago
liked a
model almost 3 years ago
liked a
Space almost 3 years ago