arXiv:2603. 16428v2 Announce Type: replace-cross Abstract: Fine-tuning Large Language Models (LLMs) has become essential for domain adaptation, but its memory-intensive property exceeds the capabilities of most GPUs.
arXiv:2607. 07862v1 Announce Type: cross Abstract: The evolution of compute infrastructure has transformed multi-GPU systems into tightly integrated shared-memory structures.
DualPace is an LLM serving runtime that decouples scheduling between a GPU engine and an AMX‑enabled CPU engine, allowing the CPU to independently batch and advance decodes while still cooperating with the GPU for dense stages. This design expands key‑value cache capacity into host memory and reduces bursty CPU progress, leading to significant throughput gains—up to 43.2 % over NEO and 90.6 % over vLLM—without increasing latency. The system also maintains low time‑to‑first‑token and supports higher request rates at comparable tail latency.
arXiv:2607. 22614v1 Announce Type: new Abstract: RL-based LLM post-training increasingly disaggregates Rollout and Training across separate GPU resources, but static GPU partitioning suffers from severe pipeline bubbles under long-tail rollout latency.
The paper investigates how a single SM utilization metric can misrepresent the true workload of large language model (LLM) inference on Nvidia Hopper GPUs. By profiling vLLM with FlashAttention‑3 and cuBLASLt on an H100 NVL across various phases (cold prefill, warm prefill, and decode) and varying sequence length and batch size, the authors replace the single utilization figure with eight detailed counter‑validated views. These views, tied to specific Nsight Compute counters or formulas, reveal how factors such as fragment fill, occupancy limits, stall signatures, wave quantization, and kernel selection create utilization gaps across four production models and six per‑layer kernel roles.
arXiv:2603. 02376v2 Announce Type: replace-cross Abstract: Computation and communication in distributed LLM training and inference are traditionally optimized in isolation; expert-crafted systems such as DeepEP, FLUX, and TokenWeave show the potential of co-design but require deep systems expertise and hardware-specific tuning; CUCo is an agentic framework that automates compute-communication co-design of CUDA kernels by combining a structured design-space formalization with a correctness-first fast-path agent for reliable baselines and an evolution-driven slow-path agent for high-performance strategies, achieving up to 1.
The paper addresses the problem of non‑deterministic outputs from large language models (LLMs) when run on different GPU architectures, caused by floating‑point non‑associativity and hardware‑dependent kernel choices. It proposes a set of fixed‑configuration fused‑upcast GEMM kernels that load 16‑bit weights, upcast to FP32, and perform IEEE‑754 compliant reductions in a problem‑shape‑dependent order, ensuring identical linear‑layer outputs across NVIDIA Ampere, Ada, and Hopper GPUs. The new approach achieves 1.17–3.1× faster end‑to‑end performance than existing solutions and halves weight‑memory traffic while maintaining cross‑architecture reproducibility.
arXiv:2607. 02521v1 Announce Type: cross Abstract: SwiGLU is the dominant MLP activation in modern large language models, yet its intermediate tensor materialization costs 9-37% of MLP execution time.
The PCIe transfer latency is silently bottlenecking your agentic inference. Here is how building a custom device-resident vector search kernel bypasses the CPU to unlock deterministic microsecond tail latencies.
arXiv:2606. 09200v1 Announce Type: cross Abstract: The rapid growth of large-scale machine learning (ML) has made distributed training across multiple GPUs a fundamental component of modern ML systems.
arXiv:2607. 19353v1 Announce Type: new Abstract: Confidential computing is becoming a practical deployment requirement for AI inference workloads that process sensitive inputs or protect proprietary model assets.