nanovdr/NanoVDR-D-Fast-Qwen3VL8B-4096
Feature Extraction • 0.5B • Updated • 2
🔄 In a Training Loop
Xinping Zhao
Yuki131
AI & ML interests
LLMs, RAG, Embedding, Reranker——A Pokémon Trainer on a journey to become a Pokémon Master.
Recent Activity
liked a model about 16 hours ago
nanovdr/NanoVDR-D-Fast-Qwen3VL8B-4096 liked a model 1 day ago
zeroentropy/zerank-2-reranker repliedto their post 2 days ago
Test-Time Scaling for Rerankers?
Can rerankers scale at test time—not by generating longer reasoning traces, but by selectively using richer document representations?
KaLM-Reranker-V1 supports Matryoshka compression from 1× to 32×, which suggests a progressive multi-fidelity pipeline:
- Embedding retrieval → Top-100
- KaLM-Reranker @ 32× compression → Top-20
- The same reranker @ 2× compression → final ranking
The intuition is simple: cheaply screen many candidates, then allocate higher-fidelity cross-attention only to the most promising ones.
For 100@32× → 20@2×, the passage-token interaction budget is roughly 31.8% of directly running 100@2×, before fixed model overheads. The key question is whether it can retain nearly the same ranking quality.
We’re considering evaluating nDCG–latency Pareto curves.
Would you consider this a useful form of test-time scaling for retrieval?
https://huggingface.co/KaLM-Embedding/KaLM-Reranker-V1-Nano
https://huggingface.co/KaLM-Embedding/KaLM-Reranker-V1-Small
https://huggingface.co/KaLM-Embedding/KaLM-Reranker-V1-Large
https://huggingface.co/papers/2606.22807
https://huggingface.co/collections/KaLM-Embedding/lychee-kalm-reranker
https://huggingface.co/KaLM-EmbeddingOrganizations
liked a
model about 16 hours ago
liked a
model 1 day ago
replied to their post 2 days ago
Great point—recall@20 after the 32× stage would clearly show what is lost during early filtering. We’ll include it alongside final nDCG and latency. 😏
upvoted an article 3 days ago
Article
Bekko Embedding: how small can a multilingual retrieval model be?
hotchpotch
• • 6 reacted to their post with 👍 3 days ago
Post
2137
Test-Time Scaling for Rerankers?
Can rerankers scale at test time—not by generating longer reasoning traces, but by selectively using richer document representations?
KaLM-Reranker-V1 supports Matryoshka compression from 1× to 32×, which suggests a progressive multi-fidelity pipeline:
- Embedding retrieval → Top-100
- KaLM-Reranker @ 32× compression → Top-20
- The same reranker @ 2× compression → final ranking
The intuition is simple: cheaply screen many candidates, then allocate higher-fidelity cross-attention only to the most promising ones.
For 100@32× → 20@2×, the passage-token interaction budget is roughly 31.8% of directly running 100@2×, before fixed model overheads. The key question is whether it can retain nearly the same ranking quality.
We’re considering evaluating nDCG–latency Pareto curves.
Would you consider this a useful form of test-time scaling for retrieval?
KaLM-Embedding/KaLM-Reranker-V1-Nano
KaLM-Embedding/KaLM-Reranker-V1-Small
KaLM-Embedding/KaLM-Reranker-V1-Large
KaLM-Reranker-V1: Fast but Not Late Interaction for Compressed Document Reranking (2606.22807)
https://huggingface.co/collections/KaLM-Embedding/lychee-kalm-reranker
KaLM-Embedding
Can rerankers scale at test time—not by generating longer reasoning traces, but by selectively using richer document representations?
KaLM-Reranker-V1 supports Matryoshka compression from 1× to 32×, which suggests a progressive multi-fidelity pipeline:
- Embedding retrieval → Top-100
- KaLM-Reranker @ 32× compression → Top-20
- The same reranker @ 2× compression → final ranking
The intuition is simple: cheaply screen many candidates, then allocate higher-fidelity cross-attention only to the most promising ones.
For 100@32× → 20@2×, the passage-token interaction budget is roughly 31.8% of directly running 100@2×, before fixed model overheads. The key question is whether it can retain nearly the same ranking quality.
We’re considering evaluating nDCG–latency Pareto curves.
Would you consider this a useful form of test-time scaling for retrieval?
KaLM-Embedding/KaLM-Reranker-V1-Nano
KaLM-Embedding/KaLM-Reranker-V1-Small
KaLM-Embedding/KaLM-Reranker-V1-Large
KaLM-Reranker-V1: Fast but Not Late Interaction for Compressed Document Reranking (2606.22807)
https://huggingface.co/collections/KaLM-Embedding/lychee-kalm-reranker
- 2 replies
commented a
paper 3 days ago
reacted to their post with 🔥 3 days ago
Post
2137
Test-Time Scaling for Rerankers?
Can rerankers scale at test time—not by generating longer reasoning traces, but by selectively using richer document representations?
KaLM-Reranker-V1 supports Matryoshka compression from 1× to 32×, which suggests a progressive multi-fidelity pipeline:
- Embedding retrieval → Top-100
- KaLM-Reranker @ 32× compression → Top-20
- The same reranker @ 2× compression → final ranking
The intuition is simple: cheaply screen many candidates, then allocate higher-fidelity cross-attention only to the most promising ones.
For 100@32× → 20@2×, the passage-token interaction budget is roughly 31.8% of directly running 100@2×, before fixed model overheads. The key question is whether it can retain nearly the same ranking quality.
We’re considering evaluating nDCG–latency Pareto curves.
Would you consider this a useful form of test-time scaling for retrieval?
KaLM-Embedding/KaLM-Reranker-V1-Nano
KaLM-Embedding/KaLM-Reranker-V1-Small
KaLM-Embedding/KaLM-Reranker-V1-Large
KaLM-Reranker-V1: Fast but Not Late Interaction for Compressed Document Reranking (2606.22807)
https://huggingface.co/collections/KaLM-Embedding/lychee-kalm-reranker
KaLM-Embedding
Can rerankers scale at test time—not by generating longer reasoning traces, but by selectively using richer document representations?
KaLM-Reranker-V1 supports Matryoshka compression from 1× to 32×, which suggests a progressive multi-fidelity pipeline:
- Embedding retrieval → Top-100
- KaLM-Reranker @ 32× compression → Top-20
- The same reranker @ 2× compression → final ranking
The intuition is simple: cheaply screen many candidates, then allocate higher-fidelity cross-attention only to the most promising ones.
For 100@32× → 20@2×, the passage-token interaction budget is roughly 31.8% of directly running 100@2×, before fixed model overheads. The key question is whether it can retain nearly the same ranking quality.
We’re considering evaluating nDCG–latency Pareto curves.
Would you consider this a useful form of test-time scaling for retrieval?
KaLM-Embedding/KaLM-Reranker-V1-Nano
KaLM-Embedding/KaLM-Reranker-V1-Small
KaLM-Embedding/KaLM-Reranker-V1-Large
KaLM-Reranker-V1: Fast but Not Late Interaction for Compressed Document Reranking (2606.22807)
https://huggingface.co/collections/KaLM-Embedding/lychee-kalm-reranker
- 2 replies
posted an update 3 days ago
Post
2137
Test-Time Scaling for Rerankers?
Can rerankers scale at test time—not by generating longer reasoning traces, but by selectively using richer document representations?
KaLM-Reranker-V1 supports Matryoshka compression from 1× to 32×, which suggests a progressive multi-fidelity pipeline:
- Embedding retrieval → Top-100
- KaLM-Reranker @ 32× compression → Top-20
- The same reranker @ 2× compression → final ranking
The intuition is simple: cheaply screen many candidates, then allocate higher-fidelity cross-attention only to the most promising ones.
For 100@32× → 20@2×, the passage-token interaction budget is roughly 31.8% of directly running 100@2×, before fixed model overheads. The key question is whether it can retain nearly the same ranking quality.
We’re considering evaluating nDCG–latency Pareto curves.
Would you consider this a useful form of test-time scaling for retrieval?
KaLM-Embedding/KaLM-Reranker-V1-Nano
KaLM-Embedding/KaLM-Reranker-V1-Small
KaLM-Embedding/KaLM-Reranker-V1-Large
KaLM-Reranker-V1: Fast but Not Late Interaction for Compressed Document Reranking (2606.22807)
https://huggingface.co/collections/KaLM-Embedding/lychee-kalm-reranker
KaLM-Embedding
Can rerankers scale at test time—not by generating longer reasoning traces, but by selectively using richer document representations?
KaLM-Reranker-V1 supports Matryoshka compression from 1× to 32×, which suggests a progressive multi-fidelity pipeline:
- Embedding retrieval → Top-100
- KaLM-Reranker @ 32× compression → Top-20
- The same reranker @ 2× compression → final ranking
The intuition is simple: cheaply screen many candidates, then allocate higher-fidelity cross-attention only to the most promising ones.
For 100@32× → 20@2×, the passage-token interaction budget is roughly 31.8% of directly running 100@2×, before fixed model overheads. The key question is whether it can retain nearly the same ranking quality.
We’re considering evaluating nDCG–latency Pareto curves.
Would you consider this a useful form of test-time scaling for retrieval?
KaLM-Embedding/KaLM-Reranker-V1-Nano
KaLM-Embedding/KaLM-Reranker-V1-Small
KaLM-Embedding/KaLM-Reranker-V1-Large
KaLM-Reranker-V1: Fast but Not Late Interaction for Compressed Document Reranking (2606.22807)
https://huggingface.co/collections/KaLM-Embedding/lychee-kalm-reranker
- 2 replies
Chinese retrieval performance appears consistently weak across T5Gemma 2 model sizes
#6 opened 6 days ago
by Yuki131
liked a
dataset 12 days ago
liked 2
models 12 days ago
liked 2
models 13 days ago
upvoted a collection 14 days ago
updated 3
models 14 days ago
upvoted a paper 15 days ago
liked a
model 16 days ago