SearcharxivSearch

arXiv subjects

Shuailong Li

Publications and source records attributed to Shuailong Li.

9 recordsLinked to original sources

Staggered Batch Scheduling: Co-optimizing Time-to-First-Token and Throughput for High-Efficiency LLM Inference

The evolution of Large Language Model (LLM) serving towards complex, distributed architectures--specifically the P/D-separated, large-scale DP+EP paradigm--introduces distinct scheduling challenges. Unlike traditional deployments where schedulers can treat instances as black boxes, DP+EP architectures exhibit high internal synchronization costs. We identify that immediate request dispatching in such systems leads to severe in-engine queuing and parallelization bubbles, degrading Time-to-First-Token (TTFT). To address this, we propose Staggered Batch Scheduling (SBS), a mechanism that deliberately buffers requests to form optimal execution batches. This temporal decoupling eliminates internal queuing bubbles without compromising throughput. Furthermore, leveraging the scheduling window created by buffering, we introduce a Load-Aware Global Allocation strategy that balances computational load across DP units for both Prefill and Decode phases. Deployed on a production H800 cluster serving Deepseek-V3, our system reduces TTFT by 30%-40% and improves throughput by 15%-20% compared to state-of-the-art immediate scheduling baselines.

cs.DC

Hierarchical Action Recognition: A Contrastive Video-Language Approach with Hierarchical Interactions

Video recognition remains an open challenge, requiring the identification of diverse content categories within videos. Mainstream approaches often perform flat classification, overlooking the intrinsic hierarchical structure relating categories. To address this, we formalize the novel task of hierarchical video recognition, and propose a video-language learning framework tailored for hierarchical recognition. Specifically, our framework encodes dependencies between hierarchical category levels, and applies a top-down constraint to filter recognition predictions. We further construct a new fine-grained dataset based on medical assessments for rehabilitation of stroke patients, serving as a challenging benchmark for hierarchical recognition. Through extensive experiments, we demonstrate the efficacy of our approach for hierarchical recognition, significantly outperforming conventional methods, especially for fine-grained subcategories. The proposed framework paves the way for hierarchical modeling in video understanding tasks, moving beyond flat categorization.

cs.CV

Light Scalars at FASER

FASER, the ForwArd Search ExpeRiment, is a currently operating experiment at the Large Hadron Collider (LHC) that can detect light long-lived particles produced in the forward region of the LHC interacting point. In this paper, we study the prospect of detecting light CP-even and CP-odd scalars at FASER and FASER 2. Considering a model-independent framework describing the most general interactions between a CP-even or CP-odd scalar and SM particles using the notation of coupling modifiers in the effective Lagrangian, we develop the general formalism for the scalar production and decay. We then analyze the FASER and FASER 2 reaches of light scalars in the large $\tanβ$ region of the Type-I two Higgs double model as a case study, in which light scalars with relatively long lifetime could be accommodated. In the two benchmark scenarios we considered, the light (pseudo)scalar decay length varies in $(10^{-8}, 10^5)$ meters. Both FASER and FASER 2 can probe a large part of the parameter space in the large $\tanβ$ region up to $10^5$, extending beyond the constraints of the other existing experiments.

hep-ph

BSM Higgs Production at a Muon Collider

The potential of the non-Standard Model heavy Higgs bosons in 2HDM at a muon collider is studied. The pair production of the non-SM Higgs bosons via the universal gauge interactions is the dominant mechanism once above the kinematic threshold. On the other hand, single Higgs boson production associated with a pair of heavy fermions is also important in the parameter region with enhanced Yukawa couplings. Both $μ^+μ^-$ annihilation channels and Vector Boson Fusion processes are considered, as well as radiative return $s$-channel production. Different types of 2HDMs can also be distinguishable for moderate and large values of $\tanβ$.

hep-ph

Exotic Higgs Decays in the Type-II 2HDMs at Current and Future pp Colliders

The exotic decay modes of non-Standard Model Higgses can serve as powerful search channels to explore the parameter space of extended Higgs sectors. In this Snowmass contribution, we illustrate this using the Two-Higgs Doublet Model (2HDM) as an example. We first review the current experimental constraints on the parameter space of a Type-II 2HDM arising from existing searches for the exotic Higgs decay mode $A/H\rightarrow HZ/AZ$. We then present the sensitivity of future colliders to discover addition Higgs bosons using the exotic decay channels $A\rightarrow HZ$, $A\rightarrow H^\pm W^\mp$ and $H^\pm\rightarrow H W^\pm$. We find that a 100 TeV collider can probe almost the entire region of the Type-II 2HDM parameter space that survives current theoretical and experimental constraints and would therefore be an ideal machine to search for heavier Higgses in hierarchical scalar sectors.

hep-ph

Heavy Higgs Bosons in 2HDM at a Muon Collider

We study the discovery potential of the non-Standard Model (SM) heavy Higgs bosons in the Two-Higgs-Doublet Models (2HDMs) at a multi-TeV muon collider and explore the discrimination power among different types of 2HDMs. We find that the pair production of the non-SM Higgs bosons via the universal gauge interactions is the dominant mechanism once above the kinematic threshold. Single Higgs boson production associated with a pair of heavy fermions could be important in the parameter region with enhanced Yukawa couplings. For both signal final states, $μ^+μ^-$ annihilation channels dominate over the vector boson fusion (VBF) processes, except at high center of mass energies where the VBF processes receive large logarithmic enhancement with the increase of energies. Single Higgs boson $s$-channel production in $μ^+μ^-$-annihilation via the radiative return can also be important for the Type-L 2HDM in the very large $\tanβ$ region, extending the kinematic reach of the heavy Higgs boson mass to the collider energy. Considering both the production and decay of non-SM Higgs bosons, signals can be identified over the Standard Model backgrounds. With the pair production channels via annihilation, 95\% C.L. exclusion reaches in the Higgs mass up to the production mass threshold of $\sqrt{s}/2$ are possible when channels with different final states are combined. Including single production modes can extended the reach further. Different types of 2HDMs can be distinguishable for moderate and large values of $\tanβ$.

hep-ph

Type-I 2HDM under the Higgs and Electroweak Precision Measurements

We explore the extent to which future precision measurements of the Standard Model (SM) observables at the proposed $Z$-factories and Higgs factories may have impacts on new physics beyond the Standard Model, as illustrated by studying the Type-I Two-Higgs-doublet model (Type-I 2HDM). We include the contributions from the heavy Higgs bosons at the tree-level and at the one-loop level in a full model-parameter space. While only small $\tanβ$ region is strongly constrained at tree level, the large $\tanβ$ region gets constrained at loop level due to $\tanβ$ enhanced tri-Higgs couplings. We perform a multiple variable global fit with non-alignment and non-degenerate masses. We find that the allowed parameter ranges could be tightly constrained by the future Higgs precision measurements, especially for small and large values of $\tanβ$. Indirect limits on the masses of heavy Higgs bosons can be obtained, which can be complementary to the direct searches of the heavy Higgs bosons at hadron colliders. We also find that the expected accuracies at the $Z$-pole and at a Higgs factory are quite complementary in constraining mass splittings of heavy Higgs bosons. The typical results are $|\cos(β-α)| < 0.05, |Δm_Φ| < 200\ {\rm GeV}$, and $\tanβ\gtrsim 0.3$. The reaches from CEPC, FCCee and ILC are also compared, for both Higgs and $Z$-pole precision measurements. Comparing to the Type-II 2HDM, the 95\% C.L. allowed range of $\cos(β-α)$ is larger, especially for large values of $\tanβ$.

hep-ph

Comparative Studies of 2HDMs under the Higgs Boson Precision Measurements

We perform comparative studies for four types of the two Higgs Doublet Models (2HDMs) under the precision measurements of the Standard Model (SM) Higgs observables at the proposed Higgs factories. We explore the discovery potential based on the hypothetical deviations in the precision data for the 2HDMs up to one-loop level. We find $5σ$ observability from the $χ^2$ fitting in a significant theory parameter space at future Higgs factories. For the Type-I 2HDM, regions with $\cos(β-α)\lesssim -0.1$ or $\cos(β-α)\gtrsim 0.08$ are discoverable at more than $5σ$ level. For the other three types of 2HDMs, the $5σ$ region is even bigger: $|\cos(β-α)|\gtrsim 0.02$ for $\tanβ\sim 1$. At small and large values of $\tanβ$, the region in $\cos(β-α)$ is further tightened. We examine the extent to which the different 2HDM theories may be distinguishable from one to the other at the $95\%$ Confidence Level with four benchmark points as case studies. We show that a large part of the parameter space of the other types of 2HDMs can be distinguished from the benchmark points of the target model. The impacts of loop corrections are found to be significant in certain parameter regions.

hep-ph

Probing Exotic Charged Higgs Decays in the Type-II 2HDM through Top Rich Signal at a Future 100 TeV pp Collider

The exotic decay modes of non-Standard Model Higgs bosons are efficient in probing the hierarchical Two Higgs Doublet Models (2HDM). In particular, the decay mode $H^\pm\to HW^\pm$ serves as a powerful channel in searching for charged Higgses. In this paper, we analyzed the reach for $H^\pm\to HW^\pm \to t\bar{t}W$ at a 100 TeV $pp$ collider, and showed that it extends the reach of the previously studied $ττW$ final states once above the top threshold. Top tagging technique is used, in combination with the boosted decision tree classifier. Almost the entire hierarchical Type-II 2HDM parameter space can be probed via the combination of all channels at low $\tanβ$ region.

hep-ph