Speculative decoding accelerates Large Language Model (LLM) inference by decoupling draft generation from target verification. While recent parallel drafters efficiently propose long token sequences in a single forward pass, they suffer from rapid acceptance decay due to a lack of inter-token dependencies. Furthermore, indiscriminately verifying these extended blocks wastes critical batch capacity on tokens with high rejection risks, severely degrading throughput in high-concurrency serving systems. We introduce DSpark, a speculative decoding framework that unifies high-throughput parallel generation with adaptive, load-aware verification. To maintain draft quality, DSpark utilizes a semi-autoregressive architecture, coupling a parallel backbone with a lightweight sequential module, to introduce intra-block dependency modeling and mitigate suffix decay. To optimize system efficiency, DSpark employs confidence-scheduled verification, dynamically tailoring the verification length for each request based on estimated prefix survival probabilities and engine-specific throughput profiles. On offline benchmarks across diverse domains, DSpark substantially improves the accepted length over state-of-the-art autoregressive and parallel drafters. When deployed within the DeepSeek-V4 serving system under live user traffic, DSpark successfully mitigates verification waste. Compared to the established production baseline (MTP-1), DSpark accelerates per-user generation speeds by 60 to 85 percent at matched throughput levels. More importantly, by preventing severe throughput degradation under strict interactivity constraints, it enables performance tiers that were previously unattainable, shifting the Pareto frontier of our serving system.
The TL1A/DR3 and IL-23 pathways exhibit well-documented synergy in driving chronic intestinal inflammation, with TL1A enhancing IFN-γ and IL-17 production in a T cell-intrinsic manner and IL-23 stabilizing the Th17 lineage. We hypothesize that simultaneously co-targeting these two non-redundant axes with a single bispecific agent will deliver superior efficacy by fundamentally reshaping the dysregulated immune landscape in conditions like inflammatory bowel disease (IBD).
We developed a fully human, symmetric 1 + 1 IgG-formatted BsAb. And the Fc portion was engineered to extend serum half-life. Binding affinity for both TL1A and IL-23p19 were determined by surface plasmon resonance (SPR). The dual functionality was assessed using cell-based reporter assays: Inhibition of TL1A-induced NF-κB activation and IL-23-induced STAT3 phosphorylation as well as IL17 secretion from PBMCs. The immune complex formation was assessed by SEC-MALS. In vivo efficacy was determined in human TL1A/IL-23 KI mice using TNBS-induced colitis model. Developability was assessed and PK profile was evaluated in FcRn transgenic mice.
The ES302 demonstrated high-affinity binding to both targets. It potently neutralized both TL1A and IL-23 functionality from in vitro assays. In animal model, the ES302 showed significantly superior efficacy over monospecific therapies, markedly reducing disease activity, histopathological scoring, and pro-inflammatory cytokines. The molecule exhibited low immunogenicity risk, and excellent developability properties, including high Tm value, low viscosity, and superior stability under stress conditions, compatible with high-concentration formulation for subcutaneous administration. Furthermore, ES302 exhibited excellent PK profile (e.g. very long in vivo half-life) in humanized FcRn mice and NHP.
ES302 is a highly differentiated antibody with strong potential for the treatment of inflammatory bowel diseases.
n/a
Therapeutic Approaches to Autoimmunity (THER)
Hongtao Lu, Jing Gao, Dawei Sun et al.· Journal of Immunology· 0 citations