12 papers
Stronger Judges Do Not See More: Deployment-Integrity Blindness in LLM Evaluation
Jianchang Su, Dawei Xiang, Yifan Zhang +1
LLM-agent evaluation reports behavioral scores but leaves the identity of the deployed runtime stack unreported. A tampered deployment that preserves evaluated outputs passes every…
Multi-Channel Mitigation of Source-Trust Shortcuts in Fact-Checking RL Agents
Jianchang Su, Yiwei Yang, Wei Zhang
Retrieval-augmented fact-checkers often receive a reliability label, such as HIGH or LOW trust, for each evidence source. These labels should adjust the model's confidence and its…
Constraints Are Graphs, Not Chains: Exact Decoding for Diffusion Language Models
Jianchang Su, Wei Zhang
Diffusion language models (dLLMs) predict masked positions in arbitrary order, but their exact constrained decoders still encode constraints as sequential languages, whose state mu…
Refreshing Less, Selecting Better: Reusing Stale Gradient Features for Efficient Influence-Based Data Selection
Jianchang Su, Yifan Zhang, Wei Zhang
Gradient-based data selection methods such as LESS score each candidate by the alignment between its gradient and a target validation gradient, and recomputing per-example gradient…
When Can First-Order Models of Fine-Tuning Bound Forgetting?
Jianchang Su, Wei Zhang
Fine-tuning a language model on new data can make it forget facts that it should keep. We ask whether measurements taken at the start of a fine-tuning run can bound, for each prote…
Characterizing and Bridging the Diagnostic Gap in eBPF Verifier Rejections
Yusheng Zheng, Zhengjie Ji, Weichen Tao +5
eBPF lets developers run custom programs inside the Linux kernel, where a verifier proves each program safe. However, when the verifier rejects a program, the unclear error makes r…
ReFlux: Reversible Compute Placement for CXL-Enabled Storage
Yanpeng Hu, Yiwei Yang, Yusheng Zheng +4
Static offload to computational storage devices proves brittle because device-side processors throttle under sustained thermal load, while opaque, vendor-specific interfaces inflat…
SAIR: Cost-Efficient Multi-Stage ML Pipeline Autoscaling via In-Context Reinforcement Learning
Jianchang Su, Yifan Zhang, Shengkai Lin +4
Multi-stage ML inference pipelines are difficult to autoscale due to heterogeneous resources, cross-stage coupling, and dynamic bottleneck migration. We present SAIR, an autoscalin…
gpu_ext: Extensible OS Policies for GPUs via eBPF
Yusheng Zheng, Tong Yu, Yiwei Yang +8
Performance in modern GPU-centric systems increasingly depends on resource management policies, including memory placement, scheduling, and observability. However, uniform policies…
PromptSculptor: Multi-Agent Based Text-to-Image Prompt Optimization
Dawei Xiang, Wenyan Xu, Kexin Chu +5
The rapid advancement of generative AI has democratized access to powerful tools such as Text-to-Image models. However, to generate high-quality images, users must still craft deta…
Selective KV-Cache Sharing to Mitigate Timing Side-Channels in LLM Inference
Kexin Chu, Zecheng Lin, Dawei Xiang +7
Global KV-cache sharing is an effective optimization for accelerating large language model (LLM) inference, yet it introduces an API-visible timing side channel that lets adversari…
CXLMemUring: A Hardware Software Co-design Paradigm for Asynchronous and Flexible Parallel CXL Memory Pool Access
Yiwei Yang, Yusheng Zheng, Kexin Chu +3
CXL-attached memory lets servers add more memory while keeping the standard load/store programming model. The main drawback is latency. CXL memory accesses are too slow for normal…