Cyril Y. He

Alan Zhao, Cyril Y. He, Wei Xu

Deployers of online LLM services usually seek to maximize cluster-wide performance given a fixed number of GPUs. Tensor parallelism (TP) is necessary to fit modern models but scales sub-linearly as the TP degree t grows, due to cross-GPU communication and non-scalable runtime work, as predicted by Amdahl's Law. Conversely, increasing t improves memory efficiency and alleviates KV-cache contention and swapping. We identify and validate an empirical optimal TP degree t_e that balances these effects. We present Albireo, a parallel inference system that raises the attainable t_e by shrinking the non-scalable portion via overlap of scheduling and I/O with compute and sequence-parallel sampling, without changing model architectures. Across models and benchmarks, Albireo achieves up to 1.9x higher throughput, 48% lower latency, 28% higher GPU utilization, and 54% lower energy than vLLM; in production it yields up to 2x higher throughput.

Alan Zhao, Cyril Y. He

The rapid adoption of large language models (LLMs) has shifted a substantial portion of inference workloads into throughput-oriented offline regimes, where fully utilizing GPU compute requires large batch sizes. However, existing deployments face a structural tension. Data parallelism (DP) scales throughput well but replicates model weights, leaving limited GPU memory for key-value (KV) cache and constraining batch size. Model parallelism reduces per-device weights, but requires fine-grained synchronization that erodes DP's independence and scheduling flexibility. We present SiDP, a memory-efficient data-parallel paradigm for offline LLM inference that treats weights as a bandwidth-backed shared resource inside a DP group. Instead of storing the full model on every GPU, SiDP organizes weights as a distributed pool: each layer is owned by a single GPU, and other replicas access its weights on demand via two complementary execution modes: a Weight-as-a-Service (WaS) mode that streams remote weights over NVLink into a small cache in the large-batch regime, and a Compute-as-a-Service (CaS) mode that ships activations to owners in the small-batch tail. Evaluated on NVIDIA H20, H200, and B200 GPUs with Qwen3-32B, Qwen2.5-72B, and Llama-3.1-70B, SiDP increases usable KV capacity by up to 1.8x under the same configurations, and converts this into up to 1.5x higher end-to-end throughput over baselines (vLLM) for offline workloads.

Qingxiu Liu, Cyril Y. He, Hanser Jiang et al.

Mixture-of-Experts (MoE) models have become a dominant paradigm for scaling large language models, but their rapidly growing parameter sizes introduce a fundamental inefficiency during inference: most expert weights remain idle in GPU memory while competing with performance-critical runtime state such as the key-value (KV) cache. Since KV cache capacity directly determines serving throughput, this mismatch leads to underutilized memory and degraded performance. In this paper, we present FluxMoE, a new MoE inference system that decouples expert parameters from persistent GPU residency. FluxMoE introduces an expert paging abstraction that treats expert weights as streamed, transient resources, materializing them on demand and evicting them immediately after use, allowing GPU memory to be preferentially allocated to throughput-critical runtime state. We implement FluxMoE atop vLLM to enable efficient MoE inference under severe memory constraints. Experimental results demonstrate that FluxMoE achieves up to 3.0$\times$ throughput gains over vLLM in memory-intensive regimes, without compromising model fidelity.