arXiv subjects

Songyu Zhang

Publications and source records attributed to Songyu Zhang.

3 recordsLinked to original sources

arXiv2026

The Second LoViF 2026 Challenge on Real-World All-in-One Image Restoration: Methods and Results

This paper presents a review of the second LoViF Challenge on Real-World All-in-One Image Restoration. The challenge aims to advance unified image restoration under diverse real-world degradation conditions, including blur, low-light, haze, rain, and snow. It provides a common benchmark for evaluating the restoration accuracy, robustness, and generalization capability of models across multiple degradation categories within a unified framework. The competition attracted 158 registered participants, and 20 teams were included in the final ranking after their submitted results were successfully reproduced and verified. This report provides a comprehensive analysis of the submitted solutions and corresponding results, highlighting recent advances in real-world all-in-one image restoration. The summarized methods and empirical findings reveal effective design strategies and establish an updated benchmark for future research in real-world low-level vision.

cs.CV↗

arXiv2026

Making MoE-based LLM Inference Resilient with Tarragon

Mixture-of-Experts (MoE) models are increasingly used to serve LLMs at scale, but failures become common as deployment scale grows. Existing systems exhibit poor failure resilience: even a single worker failure triggers a coarse-grained, service-wide restart, discarding accumulated progress and halting the entire inference pipeline during recovery--an approach clearly ill-suited for latency-sensitive, LLM services. We present Tarragon, a resilient MoE inference framework that confines the failures impact to individual workers while allowing the rest of the pipeline to continue making forward progress. Tarragon exploits the natural separation between the attention and expert computation in MoE-based transformers, treating attention workers (AWs) and expert workers (EWs) as distinct failure domains. Tarragon introduces a reconfigurable datapath to mask failures by rerouting requests to healthy workers. On top of this datapath, Tarragon implements a self-healing mechanism that relaxes the tightly synchronized execution of existing MoE frameworks. For stateful AWs, Tarragon performs asynchronous, incremental KV cache checkpointing with per-request restoration, and for stateless EWs, it leverages residual GPU memory to deploy shadow experts. These together keep recovery cost and recomputation overhead extremely low. Our evaluation shows that, compared to state-of-the-art MegaScale-Infer, Tarragon reduces failure-induced stalls by 160-213x (from ~64 s down to 0.3-0.4 s) while preserving performance when no failures occur.

cs.DC↗

arXiv2025

Palladium: A DPU-enabled Multi-Tenant Serverless Cloud over Zero-copy Multi-node RDMA Fabrics

Serverless computing promises enhanced resource efficiency and lower user costs, yet is burdened by a heavyweight, CPU-bound data plane. Prior efforts exploiting shared memory reduce overhead locally but fall short when scaling across nodes. Furthermore, serverless environments can have unpredictable and large-scale multi-tenancy, leading to contention for shared network resources. We present Palladium, a DPU-centric serverless data plane that reduces the CPU burden and enables efficient, zero-copy communication in multi-tenant serverless clouds. Despite the limited general-purpose processing capability of the DPU cores, Palladium strategically exploits the DPU's potential by (1) offloading data transmission to high-performance NIC cores via RDMA, combined with intra-node shared memory to eliminate data copies across nodes, and (2) enabling cross-processor (CPU-DPU) shared memory to eliminate redundant data movement, which overwhelms wimpy DPU cores. At the core of Palladium is the DPU-enabled network engine (DNE) -- a lightweight reverse proxy that isolates RDMA resources from tenant functions, orchestrates inter-node RDMA flows, and enforces fairness under contention. To further reduce CPU involvement, Palladium performs early HTTP/TCP-to-RDMA transport conversion at the cloud ingress, bridging the protocol mismatch before client traffic enters the RDMA fabric, thus avoiding costly protocol translation along the critical path. We show that careful selection of RDMA primitives (i.e., two-sided instead of one-sided) significantly affects the zero-copy data plane. Our preliminary experimental results show that enabling DPU offloading in Palladium improves RPS by 20.9x. The latency is reduced by a factor of 21x in the best case, all the while saving up to 7 CPU cores, and only consuming two wimpy DPU cores.

cs.NI↗