·
DataBubble
  • Home
  • Models
  • News
  • Compare
  • Boards
  • Pricing
  • About
  • Newsletter
  • Methodology
  • Contact
Latest
What Transfers from Training-Only Priors? Separating Knowledge, Ordinal Targets, and Optimization34m◆SpectrumQuant: Spectrum Alignment for LLM Quantization34m◆A Mixture-Gated Theory of Anisotropic Representation Dynamics in Transformers34m◆Privacy Accounting for Empirical Denoisers34m◆Foundations of the Scaling Law34m◆Lean Finder V2: Semantic Search for Mathlib That Disambiguates User Queries34m◆The Value of System 1 Models in Medicine34m◆Harmful Responses Are Jailbreak Templates: Transferring Autoregressive Jailbreaks to Diffusion Language Models34m◆Stochastic Inverse Problem: stability, regularization and Wasserstein gradient flow34m◆Quantum-aware Transformer model for state classification34m◆Deep Learning Warm Starts for Trajectory Optimization on the International Space Station34m◆BrainRotViT: Transformer-ResNet Hybrid for Explainable Modeling of Brain Aging from 3D sMRI34m◆MedInsightBench: Evaluating Medical Analytics Agents Through Multi-Step Insight Discovery in Multimodal Medical Data34m◆Isotropic Activation Functions Enable Deindividuated Neurons and Adaptive Topologies34m◆Half-Truths Break Similarity-Based Retrieval34m◆How do standard time series forecasting benchmarks translate to finance? A broad study of risk adjusted performance34m◆Temporal Fair Division in Multi-Agent Systems: From Precise Alternation Metrics to Scalable Coordination Proxies34m◆Causal Representation Learning for Generalisable Recommendation34m◆Instance-dependent Stochastic Lipschitz bandit34m◆Driving Video Retrieval for Complex Queries with Structured Grounding34m◆What Transfers from Training-Only Priors? Separating Knowledge, Ordinal Targets, and Optimization34m◆SpectrumQuant: Spectrum Alignment for LLM Quantization34m◆A Mixture-Gated Theory of Anisotropic Representation Dynamics in Transformers34m◆Privacy Accounting for Empirical Denoisers34m◆Foundations of the Scaling Law34m◆Lean Finder V2: Semantic Search for Mathlib That Disambiguates User Queries34m◆The Value of System 1 Models in Medicine34m◆Harmful Responses Are Jailbreak Templates: Transferring Autoregressive Jailbreaks to Diffusion Language Models34m◆Stochastic Inverse Problem: stability, regularization and Wasserstein gradient flow34m◆Quantum-aware Transformer model for state classification34m◆Deep Learning Warm Starts for Trajectory Optimization on the International Space Station34m◆BrainRotViT: Transformer-ResNet Hybrid for Explainable Modeling of Brain Aging from 3D sMRI34m◆MedInsightBench: Evaluating Medical Analytics Agents Through Multi-Step Insight Discovery in Multimodal Medical Data34m◆Isotropic Activation Functions Enable Deindividuated Neurons and Adaptive Topologies34m◆Half-Truths Break Similarity-Based Retrieval34m◆How do standard time series forecasting benchmarks translate to finance? A broad study of risk adjusted performance34m◆Temporal Fair Division in Multi-Agent Systems: From Precise Alternation Metrics to Scalable Coordination Proxies34m◆Causal Representation Learning for Generalisable Recommendation34m◆Instance-dependent Stochastic Lipschitz bandit34m◆Driving Video Retrieval for Complex Queries with Structured Grounding34m◆
News/Smarter and Cheaper at Once: Byte-Exact KV-Cache Grafting Turns a Frozen Small Model into a Verified-Knowledge Flywheel
arxiv
PublishedJuly 18, 2026 at 4:00 AM
▲bullish

Smarter and Cheaper at Once: Byte-Exact KV-Cache Grafting Turns a Frozen Small Model into a Verified-Knowledge Flywheel

Source
arxiv.orgfull article ↗
Read on arxiv→
Publisher summary· verbatim

arXiv:2607.14431v1 Announce Type: cross Abstract: We report a way to make a frozen small language model both more capable and dramatically cheaper at once, without changing any weights. Verified knowledge is deposited once as a byte-exact key-value (KV) state artifact and later restored, by graft, i

Stay posted· Newsletter

Three short emails a week: Monday, Wednesday and Friday. Top movers, new models and the most-referenced story.

// no spam · unsubscribe one-click · free forever

Your email is used only to send The Bubble Brief. Privacy policy

Discussion
Mentioned models
01
  • 01
    Gemma-4-12B
Source
↗
arxiv
Read original ↗All from arxiv →
Tags
03
#language-model#optimization#performance

No replies yet. Be first.

Mentioned models
01
  • 01
    Gemma-4-12B
Source
↗
arxiv
Read original ↗All from arxiv →
Tags
03
#language-model#optimization#performance

Related coverage

More from ARXIV
arxivWhat Transfers from Training-Only Priors? Separating Knowledge, Ordinal Targets, and Optimization34marxivSpectrumQuant: Spectrum Alignment for LLM Quantization34marxivA Mixture-Gated Theory of Anisotropic Representation Dynamics in Transformers34marxivPrivacy Accounting for Empirical Denoisers34m
The Bubble Brief
MON · WED · FRI

Read language-model insights every Monday, Wednesday and Friday — top movers, new models, story of the week.

// no spam · unsubscribe one-click · free forever

Your email is used only to send The Bubble Brief. Privacy policy

Originally published on arxiv ↗
PrivacyTermsRefundsLegal notice|AboutMethodologyContact

© 2026 Databubble

Built by Marouane Gazouzi · Readyo, AI interview coach

HomeModelsNews