This paper proposes a preconditioning layer for stable weight conditioning in LLM training.

Turning off automatic NUMA balancing in simulation of large-scale spiking networks reduces energy consumption by 30%.

This paper introduces DGNA, a methodology to unveil the Non-Uniform Memory Access (NUMA) architecture of GPU memory hierarchy through microbenchmarking and data analysis.

The paper investigates whether using fine-grained, tensorized adapters (CP components) instead of standard LoRA ranks improves the accuracy-budget trade-off in PEFT, finding that while they fill budge…

The paper proposes a unified hybrid framework that combines data-level and algorithm-level balancing to effectively address the challenge of imbalanced regression, significantly improving predictive p…

The paper introduces Spandana, an architecture that decouples SLO enforcement from cost optimization in cloud-based online services, achieving high utilization, strict SLO adherence, and cost savings.

The paper presents a solution to enable multiple double arithmetic on NVIDIA A100 tensor cores, which are unsuited for branching operations caused by renormalization.

A simple online algorithm is presented for the stacking problem to avoid shifts with a sufficient condition involving stacking area dimension, load/unload points, and maximum items.

This paper introduces Stale Synchronous Parallel mode of execution for parallel sparse triangular linear system solve and presents a scheduler that achieves geometric-mean speed-ups of 7-30% over Grow…

Hawkeye is a system that allows perfect, precision-preserving reproduction of GPU-level matrix multiplication operations on a CPU, enabling efficient and trustworthy third-party auditing of machine le…

The paper introduces NeWTral, a framework that restores safety alignment to specialized LLM adapters without sacrificing their domain-specific knowledge, achieving a significant reduction in attack su…

This paper compares and optimizes CUDA strategies for a shallow neural network, achieving a 1.41x speedup on a large dataset.

This paper introduces DDC, a Dead-Direction Conditioner that keeps a deep network's optimization on the symmetry quotient by conditioning the optimizer's state in the orbit decomposition of a $G$-inva…

This paper presents Festina, a profiling-guided, power-aware control plane for minimizing energy consumption in serverless large language model (LLM) serving.

This paper presents a new modular architecture for NeoSLAM using modern frameworks, achieving real-time execution and minimal data discarding. It also compares NeoSLAM and RatSLAM across three dataset…

A two-layer framework using a large language model for force-conditioned reinforce learning with recovery maneuvers and force signatures.