🏝️ On Vacation

Amber

AmberLJC

6 13 13

https://amberljc.github.io/

AI & ML interests

AI4S, LLM Systems

Recent Activity

upvoted an article 4 days ago

What We Learned by Reproducing 2,200 papers from ICML

liked a dataset 4 days ago

AgentNativeResearchLab/arc-agi3-kimi-k2.7-g50t

reacted to Yuki131's post with 🔥 7 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-Embedding

View all activity

Organizations

upvoted an article 4 days ago

Article

What We Learned by Reproducing 2,200 papers from ICML

abidlabs

•

5 days ago

• 75

liked a dataset 4 days ago

AgentNativeResearchLab/arc-agi3-kimi-k2.7-g50t

Updated 7 days ago • 2.84k • 2

reacted to Yuki131's post with 🔥 7 days ago

Post

2161

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