arXiv Machine Learning
Aug 27
How Much Rank Does LoRA Need? Rank-Error Bounds for Transformer Attention
arXiv:2608. 26052v1 Announce Type: new Abstract: Choosing the rank of a low-rank adaptation (LoRA) update is usually an empirical task.
When Can Prefixes Compile LoRA? Exact Resource-Capped Tests for Frozen Attention
Related stories
arXiv AI
Jun 2
Finer Parameter Steps for Low-Rank PEFT: A Controlled Study with CP Tensor Adapters
arXiv:2606. 00428v1 Announce Type: cross Abstract: Low-rank adapters are usually compared by sweeping a small set of ranks, but the rank also fixes the resolution of the parameter budget.
arXiv Machine Learning
Sep 23
Intrinsic Interaction Geometry Controls the Low-Rank Complexity of Softmax Attention
arXiv:2608.28150v2 Announce Type: replace Abstract: How much matrix rank is required to preserve every bounded value output of normalized softmax attention? We study the unrestricted maximum-row-\(\e...
arXiv Computation and Language
Sep 15
Prefix Sharing Is a Sorting Problem
arXiv:2609.13692v1 Announce Type: cross Abstract: LLM serving reuses KV cache by exact prefix match, so when a prompt is assembled from a set of reusable pieces -- retrieved passages, tool definition...
arXiv AI
Jul 23
MoA-Structured Decode Attention DNF Derivation, KV-Cache Accumulation, GQA/MQA, and OpenACC Kernel
arXiv:2607. 19456v1 Announce Type: cross Abstract: We derive four memory-optimal inference artifacts for transformer attention using the Mathematics of Arrays (MoA), each following directly from the forward-pass Denotational Normal Form (DNF) of with the query-row index fixed to the current decode step.
arXiv AI
Jul 22
CompilerKV: Risk-Adaptive KV Compression via Offline Experience Compilation
arXiv:2602. 08686v3 Announce Type: replace-cross Abstract: Prefill-only KV compression freezes a token subset at the end of prefill and decodes from it without further eviction.
arXiv Machine Learning
Jun 3
Compress then Merge: From Multiple LoRAs into One Low-Rank Adapter
arXiv:2606. 03723v1 Announce Type: new Abstract: Low-rank adaptation (LoRA) enables parameter-efficient specialization of foundation models, but the proliferation of task-specific adapters fragments capabilities across many adapters, complicating reuse and deployment.
arXiv AI
Aug 24
From Attention Masks to Inert Zero-Vector Tokens: OAttention and O-Closure for Token Dynamics
The paper introduces OAttention, a token‑level attention mechanism that assigns each token a presence coefficient based on its hidden representation. This coefficient both gates the token’s output and weights its contribution to other tokens, making zero‑vector tokens behave as true zeros and enabling exact null‑receiver, null‑source, and empty‑support properties. The authors extend this idea to local O‑components and an O‑Transformer, and demonstrate small performance changes when retrofitting a pretrained TabPFN model.
arXiv Machine Learning
Sep 22
ValueDiff: Value-Geometric KV Cache Eviction for Sink-Suppressed LLMs
ValueDiff introduces a value‑geometric KV cache eviction strategy for large language models that suppress attention sinks. It ranks tokens by the L2 deviation of their value vectors from the cache mean, a score that aligns with minimal‑disturbance eviction under a max‑entropy assumption. Across several benchmarks—RULER, LongBench, and MATH‑500—ValueDiff consistently retains a higher proportion of useful tokens than prior methods, especially under tight cache budgets.
arXiv Computation and Language
Sep 10
Better Together: Complementary Query Rewriting Under a Strong RAG Baseline
arXiv:2609.05637v2 Announce Type: replace Abstract: A popular way to improve Retrieval-Augmented Generation (RAG) is to rewrite the user's question into several variants and search with all of them....
arXiv Machine Learning
Sep 18
A Table-Free Index for Tapered Memoization Grids: Compact Out-of-Core Evaluation of Functions of Sorted Arguments
The paper presents a table‑free index for tapered memoization grids, enabling compact out‑of‑core evaluation of functions that depend on sorted arguments. By showing that the grid’s key set corresponds to multiset combinations, the authors derive a closed‑form O(d) ranking and unranking scheme that removes the need for large preprocessing tables and allows order‑free parallel construction. The resulting values‑only flat array uses significantly less memory than hash‑map memoization, offers faster query times once cache limits are exceeded, and remains operable with memory‑mapped storage beyond RAM.
arXiv AI
Sep 12
SemVerBench: Benchmarking LLM Comprehension of Version-Constraint Resolution Semantics
SemVerBench is a benchmark that evaluates how well large language models (LLMs) understand and apply version-constraint resolution semantics, such as determining whether a version satisfies constraints like ^1.2.3 or >=2.0. The study finds that many models struggle with certain corner cases, with GPT‑5.1 performing poorly while Claude and Opus perform much better. The authors suggest that the failures stem from an activation/application gap rather than a lack of knowledge, and recommend that coding agents delegate version resolution to a dedicated resolver tool.