2027
SwarmPilot: Agentic Scheduling for Agentic Systems
Yeqi Huang*, Yanwei Ye*, Guomin Chen, Wenhao Su, Bin Gong, Jialian Li, Zhan Lu, Yangshen Deng, Xuan Sun, Le Xu, Luo Mai (* equal contribution)
EuroSys '27 Spring 2027 Accepted
SwarmPilot introduces agentic scheduling for low-latency agentic applications. It combines scheduling-specific predictors over prompt, device, runtime, and target-model features with a scheduler–agent framework, reducing tail latency by up to 61.5% and sustaining up to 2× the throughput of baseline schedulers under the same SLO in the reported evaluation.
SwarmPilot: Agentic Scheduling for Agentic Systems
Yeqi Huang*, Yanwei Ye*, Guomin Chen, Wenhao Su, Bin Gong, Jialian Li, Zhan Lu, Yangshen Deng, Xuan Sun, Le Xu, Luo Mai (* equal contribution)
EuroSys '27 Spring 2027 Accepted
SwarmPilot introduces agentic scheduling for low-latency agentic applications. It combines scheduling-specific predictors over prompt, device, runtime, and target-model features with a scheduler–agent framework, reducing tail latency by up to 61.5% and sustaining up to 2× the throughput of baseline schedulers under the same SLO in the reported evaluation.
2026
Wavel: A Fast and Efficient Compilation System for Wafer-Scale Accelerators
Yeqi Huang, Congjie He*, Haocheng Xiao*, Yanwei Ye*, Yi-Chieh Wang*, Boyao Song*, Yangshen Deng*, Ziming Miao*, Lingxiao Ma*, Fan Yang*, Luo Mai* (* equal contribution)
SOSP 2026 2026 Accepted
Wavel presents a compilation system for wafer-scale accelerators that makes executable schedules explicit, uses locality-preserving CUT and TUNE primitives, and evaluates physical cost during search. The reported results show higher throughput than expert-optimized WaferLLM and substantial speedups over a Full-Chip baseline on WSE-3 workloads.
Wavel: A Fast and Efficient Compilation System for Wafer-Scale Accelerators
Yeqi Huang, Congjie He*, Haocheng Xiao*, Yanwei Ye*, Yi-Chieh Wang*, Boyao Song*, Yangshen Deng*, Ziming Miao*, Lingxiao Ma*, Fan Yang*, Luo Mai* (* equal contribution)
SOSP 2026 2026 Accepted
Wavel presents a compilation system for wafer-scale accelerators that makes executable schedules explicit, uses locality-preserving CUT and TUNE primitives, and evaluates physical cost during search. The reported results show higher throughput than expert-optimized WaferLLM and substantial speedups over a Full-Chip baseline on WSE-3 workloads.
Ryze: Evidence-Enriched Data Synthesis from Biomedical Papers
Yeqi Huang, Yue Chen, Yanwei Ye, Guanhao Su, Luo Mai
ACL 2026 Demo 2026 Accepted
General-purpose VLMs remain unreliable for biomedical research because valid answers in scientific papers depend on evidence split across figures, tables, charts, captions, and referring text. Existing post-training pipelines are bottlenecked by costly expert annotation and by synthetic data that drops this evidence structure. We present Ryze, a fully automated system that converts raw biomedical papers into an evidence-enriched training set and a domain-specialized VLM. Ryze synthesizes QA pairs with complete supporting evidence (visual element, caption, extracted structure, and referring paragraphs), reduces layout and OCR errors via chart/table-aware extraction and LLM-based cleansing, and applies a two-stage post-training strategy combining supervised fine-tuning with reinforcement learning. Starting from Qwen3-VL-8B, Ryze produces BioVLM-8B at under $200, achieving 48.0% weighted accuracy on LAB-Bench—outperforming the base model by +12.6% and surpassing GPT-5.2 by +3.8%. We release Ryze as open source together with the trained BioVLM-8B model.
Ryze: Evidence-Enriched Data Synthesis from Biomedical Papers
Yeqi Huang, Yue Chen, Yanwei Ye, Guanhao Su, Luo Mai
ACL 2026 Demo 2026 Accepted
General-purpose VLMs remain unreliable for biomedical research because valid answers in scientific papers depend on evidence split across figures, tables, charts, captions, and referring text. Existing post-training pipelines are bottlenecked by costly expert annotation and by synthetic data that drops this evidence structure. We present Ryze, a fully automated system that converts raw biomedical papers into an evidence-enriched training set and a domain-specialized VLM. Ryze synthesizes QA pairs with complete supporting evidence (visual element, caption, extracted structure, and referring paragraphs), reduces layout and OCR errors via chart/table-aware extraction and LLM-based cleansing, and applies a two-stage post-training strategy combining supervised fine-tuning with reinforcement learning. Starting from Qwen3-VL-8B, Ryze produces BioVLM-8B at under $200, achieving 48.0% weighted accuracy on LAB-Bench—outperforming the base model by +12.6% and surpassing GPT-5.2 by +3.8%. We release Ryze as open source together with the trained BioVLM-8B model.