arXiv:2607. 01984v1 Announce Type: cross Abstract: Newer lightweight convolutional neural networks are often presented as improving predictive performance and deployment efficiency, but such claims require controlled evaluation.
arXiv:2608. 20210v1 Announce Type: cross Abstract: Small language models are usually built like large ones and then squeezed onto a CPU afterwards.
arXiv:2609.21058v1 Announce Type: cross Abstract: Language models can now write GPU kernels that outperform PyTorch. We evaluate five model configurations on KernelBench level 1 and find that a front...
arXiv:2609.14247v1 Announce Type: cross Abstract: Bird-strike mitigation at wind farms requires identifying distant birds that span only tens of pixels, so the classifier's input resolution N is a de...
arXiv:2609.40127v1 Announce Type: cross Abstract: Modern transformers pair impressive capabilities with substantial memory and compute demands. Low-rank weight factorization reduces both while keepin...
The study re‑examines a reported advantage of a routed ternary (1.58‑bit) language model over a full‑precision transformer at 60K parameters. By running controlled experiments with multiple seeds and a fixed training recipe, the authors find that the apparent benefit largely stems from the choice of baseline model shape rather than the ternary architecture itself. While the routed model does outperform other shapes at a larger 130M‑byte budget, its advantage diminishes when a plain gated diagonal‑SSM block is used, and the ternary penalty varies with architecture and quantization details.
Small language models are usually built like large ones and then squeezed onto a CPU afterwards. We did the opposite: we fixed the target first, one user, one token at a time, 4-bit weights, ordinary CPU, and chose the architecture to suit it.
The paper presents a redesign of the LiSenNet speech‑enhancement model for deployment on the STM32N6570‑DK Neural‑ART microcontroller accelerator. By replacing the recurrent bottleneck with convolutional mixers, converting unsupported operations to static int8 primitives, and using bounded decoder activations, the authors achieve an NPU‑compatible model that matches or surpasses the original LiSenNet in quality (PESQ 3.08 vs 3.01 FP32) while running each 16 ms input hop in 4.83 ms (real‑time factor 0.30). The study demonstrates that co‑designing parameter count, operator compatibility, quantization range, and streaming state is essential for efficient real‑time speech enhancement on constrained NPUs.
arXiv:2607. 05872v1 Announce Type: new Abstract: Memory-efficient optimizers such as GaLore train large language models by projecting gradients onto a rank-r subspace recomputed every T steps, assuming this subspace is a slowly drifting object that can be tracked.
arXiv:2606. 24780v1 Announce Type: new Abstract: Progress in deep learning is, at scale, more a matter of systems engineering than of modelling: the behaviour of a model in training (its throughput, its memory footprint, and the numerical fidelity of the result) is determined less by the architecture itself than by how that architecture is expressed on the hardware.
The paper investigates whether per‑frame early exit can improve compute‑matched performance for on‑device speech enhancement. By supervising every intermediate depth of a causal model and fine‑tuning output heads, the authors produce a family of static models that are more Pareto‑efficient than those trained from scratch, achieving up to 0.11 higher PESQ for equivalent compute and matching the best PESQ at 30% less compute. After int8 quantization, the dynamic enhancer performs on the same latency‑quality frontier as static models on an STM32N6 microcontroller, with the policy execution adding only 26 µs per frame and a 2.2% latency overhead from graph splitting.
Task-Aware Spectral Pruning (TASP) is a post‑training framework that tailors sparse masks to specific tasks by calibrating module‑level spectral descriptors against task‑specific ablation effects. It constructs masks that close grouped‑query‑attention and SwiGLU dependencies, routing each user turn to a single compiled mask that remains fixed during prefill and decoding. In experiments, TASP achieves a 43% active‑FLOP reduction while preserving 97.7% of the dense BF16 performance on Llama‑3‑70B, and delivers a 1.44× speedup on an A100 80GB with INT8‑weight/BF16‑compute, reducing decode latency from 45.2 to 31.3 ms/token.