arXiv:2606. 15004v1 Announce Type: cross Abstract: Deploying neural networks on low-power microcontrollers (MCUs) requires selecting model architectures under tight memory, latency, and energy constraints.
By Joseph Q. Zales, Pragya Sharma, Mani Srivastava
MCUFlowNet introduces a hardware‑aware neural architecture search framework, PACE‑NAS, that separates memory usage from latency for optical flow networks on microcontrollers. By fixing the backbone’s peak SRAM and searching for efficient operators and depths, the authors reduce peak SRAM usage by 18% and improve accuracy with lightweight context modules. The resulting MCUFlowNet models, particularly MCUFlowNet‑S, run 1.76× faster than a retrained EdgeFlowNet while achieving comparable or better performance in simulated drone flight tests, and the approach also benefits semantic segmentation tasks.
By Enmin Lin, Jona Beysens
The paper introduces Primus Projection, a hybrid engine that combines GPU microbenchmarks with analytical scaling to predict multi‑node distributed training performance, and Primus Tuning Agent, an automated search that uses the projection engine as a fast oracle to explore the vast configuration space of parallelism and micro‑batching. Together, they reduce the need for costly trial‑and‑error on hundreds of GPUs, achieving predictions within ~10% of real throughput and enabling significant throughput gains—up to +27% in a Mixtral 8x22B case study—within minutes of single‑node exploration.
By Anshu Raina, Peyman Razaghi, Yuankai Chen, Cheng Yao, Yao Fu, Vidushi Goyal, Devang Patel, Wen Xie, Zhenyu Gu, Emad Barsoum
arXiv:2605. 16138v3 Announce Type: replace Abstract: Neural architecture search (NAS) is a powerful approach for automating model design, but existing methods often optimize for accuracy alone or rely on proxy metrics such as bit operations (BOPs) that correlate poorly with hardware cost.
By Jason Weitz, Dmitri Demler, Benjamin Hawks, Aaron Wang, Nhan Tran, Javier Duarte
Configuring distributed LLM training across hundreds of GPUs requires jointly optimizing tensor, pipeline, expert, context, and data parallelism alongside micro-batch size, recomputation, sharding, an...
arXiv:2605. 16138v2 Announce Type: replace Abstract: Neural architecture search (NAS) is a powerful approach for automating model design, but existing methods often optimize for accuracy alone or rely on proxy metrics such as bit operations (BOPs) that correlate poorly with hardware cost.
By Jason Weitz, Dmitri Demler, Benjamin Hawks, Aaron Wang, Nhan Tran, Javier Duarte
EdgeCIM is a hardware-software co-design framework that accelerates autoregressive decoding of small language models on edge devices. It combines a CIM macro built in 65 nm with a tile-based mapping strategy to balance pipeline stages and reduce DRAM bandwidth bottlenecks. Benchmarks show that, at INT4 precision, EdgeCIM delivers up to 7.07× higher throughput than competing edge accelerators and averages 336.4 tokens/s with 173 tokens/J across a range of models up to 4 B parameters.
By Jinane Bazzi, Mariam Rakka, Fadi Kurdahi, Mohammed E. Fouda, Ahmed Eltawil
arXiv:2604. 26968v2 Announce Type: replace-cross Abstract: Key-value (KV) cache memory management is the primary bottleneck limiting throughput and cost-efficiency in large-scale GPU inference serving.
By Sanjeev Rao Ganjihal
arXiv:2608. 08961v1 Announce Type: new Abstract: AI training's rising resource intensity is straining electricity supplies and carbon budgets, motivating systematic study of memory-efficient training on constrained hardware.
By Sarthak Mahapatra, Zihan Zhou, Khatoon Khedri, Mehdi Hosseinzadeh, Reza Rawassizadeh
arXiv:2606. 10294v1 Announce Type: cross Abstract: Deploying neural networks on unconventional hardware demands architectures that co-optimize task accuracy and platform-specific constraints such as energy cost, physical non-idealities, and numerical precision.
By Tyler King, Timothee Leleu
arXiv:2606. 28565v1 Announce Type: cross Abstract: As large language models (LLMs) move into production serving, practitioners must rapidly evaluate inference performance across diverse hardware, models, and serving parameters to meet cost and latency targets.
By Xiteng Yao, Taeho Kim, Hengzhi Pei, Xinle Liu, Kyle Ulrich, Leonard Lausen, Ashish Khetan, Xiang Song, George Karypis, Martin Herbordt
The paper introduces the Tri‑Metric Router, a deterministic, training‑free policy that chooses among Raw, Neural, and Lexical pipelines for retrieval‑augmented generation on commodity GPUs. It uses three CPU‑side signals—spatial complexity, syntactic density, and type‑token ratio—to balance VRAM headroom and latency, calibrated on LongBench qasper. The method eliminates out‑of‑memory failures and improves alignment and F1 scores compared to always‑on lexical compression without extra VRAM or training costs.
By Saipraveen Vabbilisetty, Ajay Kumar Boddepalli, Deep Narayan Mishra, Shashank Kapadia, Haoan Wang, Anupriya Sharma