The Inference Report

GPU acceleration is now routine across robotics, but cloud-hosted GPU continuous integration (CI) runners are expensive, resulting in severe under-testing of GPU-accelerated code. We present pytest-gpu-proof, an open-source pytest plugin offering a practical middle ground. Tests can be run on a local machine, signed with a receipt of exactly what ran and what it produced, and integrated into standard CPU CI workflows (e.g., GitHub Actions). The tool is open source and on PyPI, and we are actively integrating it across our lab's software stack.

Long-running LLM applications repeatedly send growing context, making prefix caching critical for reducing prefill cost. Yet prefix-cache behavior under agentic workloads remains poorly understood. We study production traces from two companies and evaluate 14 eviction algorithms across HBM-constrained and large memory-pool settings. Despite a large gap to Belady, sophisticated policies designed for traditional caches provide little benefit over LRU. The reason is structural: prefix reuse is dominated by the regular pacing of active sessions, making recency unusually predictive. Prefix caching nevertheless introduces new challenges, including heavy-tailed session footprints and highly variable miss costs as attention computation grows with sequence length. We introduce the compute-savings ratio and two offline oracles to quantify these effects. Our results show that effective prefix-cache management should retain recency as its foundation while selectively adding quick demotion for one-hit prefixes, compute-aware partial eviction for expensive misses, and capacity-dependent eviction granularity. We will release the traces and simulator to support future research.

Training with a fixed global batch limits how many distributed clients can provide examples in any one step. We examine a way to use additional server workers without increasing the batch processed by an individual workload. Global Clustered Parallel Split Learning (GCPSL) assigns clients to fixed clusters, executes a Parallel Split Learning with Global Sampling (GPSL) workload for each cluster concurrently, and periodically fuses the client and server model segments. In simulations with 256 logical clients, dividing the population across more workloads improves direct data participation, while smaller clusters can incur an accuracy cost. A four-H100 implementation of label-aware GCPSL reaches 85% CIFAR-10 validation accuracy in $6.13 \pm 0.15$ minutes over three matched runs, versus $19.09 \pm 0.45$ minutes when the same workloads are serialized. Within the four-GPU allocation, size-balanced and random fixed affiliations reach the target in similar mean times (5.70 and 5.66 minutes); size balancing increases direct participation by 3.25 percentage points. These measurements characterize a trade-off among execution concurrency, assignment information, participation, and accuracy for stable-client split learning.

Deep learning inference and training performance depends critically on GPU kernel efficiency. Modern compilers such as PyTorch Inductor automatically generate GPU kernels from high-level model code, but frequently underperform expert-written implementations by wide margins. Recent LLM-assisted kernel optimizers can close this gap for standalone kernels, yet treat compiled models as black boxes, generally optimizing individual standalone kernels without respecting the compiler's structural decisions or verifying the model end-to-end. We present KernelOPT, a multi-agent system that treats compiled models as structured artifacts. It preserves vendor library calls (cuBLAS, cuDNN) and exclusively targets generated Triton sub-kernels using five profiling-guided LLM agents. A four-gate verification cascade of static validation, multi-seed correctness, model-level float64-fallback verification, and performance gating filters candidates during optimization and verifies the re-stitched model end-to-end. If no candidate passes all four gates, the system preserves the compiler baseline. The system accepts PyTorch nn.Modules, standalone Triton kernels, and Helion kernels. Evaluated on 250 KernelBench problems, KernelOPT achieves geometric mean speedups over \texttt{torch.compile} of 1.40$\times$ (Level 1: 51/100), 1.15$\times$ (Level 2: 31/100), and 1.07$\times$ (Level 3: 12/50) across all problems.

Multi-GPU servers have become the standard building block of modern data centers, providing aggregated capacity through high-bandwidth interconnects. At the same time, workloads such as LLM inference exhibit highly dynamic memory demands, which can cause one GPU to exhaust its local memory while others remain underutilized. This mismatch motivates a model of elastic resource sharing across GPUs. We present EMA, a memory sharing system that allows GPUs within a server to borrow and reclaim memory from each other, forming an elastic pool of capacity. EMA ensures performance transparency for both borrowers and lenders. For borrowers, prefetching hides remote access costs so that applications experience remote and local memory as indistinguishable in performance. For lenders, borrowed resources remain reclaimable on demand, guaranteeing that performance never falls below that of static partitioning. While our design focuses on memory, the same principle naturally extends to other GPU resources. Our evaluation shows that EMA improves individual user throughput by up to 52%, achieves 96% of the throughput of a system provisioned with 2X capacity, and maintains latency similar to the static local baseline.

As serving capacity demand surpasses that of training, serving efficiency becomes increasingly important. Prefill-decode (P/D) disaggregation improves serving efficiency through specialization and isolation of the two phases. These benefits rest on a static partitioning. Phase demand, however, is not static. We observe that in a large LLM fleet the ratio of uncached input to output tokens has peak-to-mean ratios up to 4.7x at minute timescales, and that in a public agentic trace the hourly ratio spans a median 24.5x within a single day, while reassigning a replica takes tens of minutes. Agentic traffic sharpens the mismatch. Sizing each pool at its ninety-fifth percentile leaves up to 17% of cluster capacity unused; sizing below it converts the same imbalance into queueing and unrealized throughput. We present Crossflow, which makes this boundary elastic without changing node roles. Each decode node publishes a short-lived, revocable lease that bounds local-prefill compute, KV capacity, transfer work, and projected output. Across public and internal traces, Crossflow improves token throughput by 16.2-17.4% on geometric mean over static P/D, and by up to 43.4% at high load, while reducing mean TTFT at every evaluated point.