arXiv subjects

Yicheng Feng

Publications and source records attributed to Yicheng Feng.

At least 37 records · Page 2Linked to original sources

arXiv2025

Frontier: Simulating the Next Generation of LLM Inference Systems

Large Language Model (LLM) inference is growing increasingly complex with the rise of Mixture-of-Experts (MoE) models and disaggregated architectures that decouple components like prefill/decode (PD) or attention/FFN (AF) for heterogeneous scaling. Existing simulators, architected for co-located, dense models, are unable to capture the intricate system dynamics of these emerging paradigms. We present Frontier, a high-fidelity simulator designed from the ground up for this new landscape. Frontier introduces a unified framework to model both co-located and disaggregated systems, providing native support for MoE inference with expert parallelism (EP). It enables the simulation of complex workflows like cross-cluster expert routing and advanced pipelining strategies for latency hiding. To ensure fidelity and usability, Frontier incorporates refined operator models for improved accuracy. Frontier empowers the community to design and optimize the future of LLM inference at scale.

cs.LG↗

arXiv2025

Experimental Search for the Chiral Magnetic Effect in Relativistic Heavy-Ion Collisions: A Perspective

The chiral magnetic effect (CME) refers to generation of the electric current along a magnetic field in a chirally imbalanced system of quarks. The latter is predicted by quantum chromodynamics to arise from quark interaction with non-trivial topological fluctuations of the vacuum gluonic field. The CME has been actively searched for in relativistic heavy-ion collisions, where such gluonic field fluctuations and a strong magnetic field are believed to be present. The CME-sensitive observables are unfortunately subject to a possibly large non-CME background, and firm conclusions on a CME observation have not yet been reached. In this perspective, we review the experimental status and progress in the CME search, from the initial measurements more than a decade ago, to the dedicated program of isobar collisions in 2018 and the release of the isobar blind analysis result in 2022, to intriguing hints of a possible CME signal in Au+Au collisions, and discuss future prospects of a potential CME discovery in the anticipated high-statistic Au+Au collision data at the Relativistic Heavy-Ion Collider by 2025. We hope such a perspective will help sharpening our focus on the fundamental physics of the CME and steer its experimental search.

nucl-ex↗

arXiv2025

Being-H0: Vision-Language-Action Pretraining from Large-Scale Human Videos

We introduce Being-H0, a dexterous Vision-Language-Action model (VLA) trained on large-scale human videos. Existing VLAs struggle with complex manipulation tasks requiring high dexterity and generalize poorly to novel scenarios and tasks, primarily due to their reliance on synthetic data with significant sim-to-real gaps or teleoperated demonstrations lacking scale and diversity. To address this data bottleneck, we propose leveraging human hands as a foundation manipulator, capitalizing on the rich dexterity and scalability present in web data. Our approach centers on physical instruction tuning, a novel training paradigm that combines large-scale VLA pretraining from human videos, physical space alignment for 3D reasoning, and post-training adaptation for robotic tasks. Additionally, we introduce a part-level motion tokenization method which achieves millimeter-level reconstruction accuracy to model precise hand trajectories for action learning. To support our proposed paradigm, we further develop a comprehensive data curation pipeline that integrates heterogeneous sources -- including motion capture, VR, and RGB-only videos -- into a large-scale dataset with millions of motion-based instructional instances. We empirically show the excellence of Being-H0 in hand motion generation and instruction following, and it also scales well with model and data sizes. Importantly, we observe the expected gains of Being-H0 in real-world robotic manipulation as physical instruction tuning is applied. More details are available at https://beingbeyond.github.io/Being-H0.

cs.CV↗

arXiv2025

Unified Multimodal Understanding via Byte-Pair Visual Encoding

Multimodal large language models (MLLMs) have made significant progress in vision-language understanding, yet effectively aligning different modalities remains a fundamental challenge. We present a framework that unifies multimodal understanding by applying byte-pair encoding to visual tokens. Unlike conventional approaches that rely on modality-specific encoders, our method directly incorporates structural information into visual tokens, mirroring successful tokenization strategies in text-only language models. We introduce a priority-guided encoding scheme that considers both frequency and spatial consistency, coupled with a multi-stage training procedure based on curriculum-driven data composition. These enhancements enable the transformer model to better capture cross-modal relationships and reason with visual information. Comprehensive experiments demonstrate improved performance across diverse vision-language tasks. By bridging the gap between visual and textual representations, our approach contributes to the advancement of more capable and efficient multimodal foundation models.

cs.CV↗

arXiv2025

Being-0: A Humanoid Robotic Agent with Vision-Language Models and Modular Skills

Building autonomous robotic agents capable of achieving human-level performance in real-world embodied tasks is an ultimate goal in humanoid robot research. Recent advances have made significant progress in high-level cognition with Foundation Models (FMs) and low-level skill development for humanoid robots. However, directly combining these components often results in poor robustness and efficiency due to compounding errors in long-horizon tasks and the varied latency of different modules. We introduce Being-0, a hierarchical agent framework that integrates an FM with a modular skill library. The FM handles high-level cognitive tasks such as instruction understanding, task planning, and reasoning, while the skill library provides stable locomotion and dexterous manipulation for low-level control. To bridge the gap between these levels, we propose a novel Connector module, powered by a lightweight vision-language model (VLM). The Connector enhances the FM's embodied capabilities by translating language-based plans into actionable skill commands and dynamically coordinating locomotion and manipulation to improve task success. With all components, except the FM, deployable on low-cost onboard computation devices, Being-0 achieves efficient, real-time performance on a full-sized humanoid robot equipped with dexterous hands and active vision. Extensive experiments in large indoor environments demonstrate Being-0's effectiveness in solving complex, long-horizon tasks that require challenging navigation and manipulation subtasks. For further details and videos, visit https://beingbeyond.github.io/Being-0.

cs.RO↗

arXiv2025

VideoOrion: Tokenizing Object Dynamics in Videos

We present VideoOrion, a Video Large Language Model (Video-LLM) that explicitly captures the key semantic information in videos - the spatial-temporal dynamics of objects throughout the videos. VideoOrion employs expert vision models to extract object dynamics through a detect-segment-track pipeline, encoding them into a set of object tokens by aggregating spatial-temporal object features. Our method addresses the persistent challenge in Video-LLMs of efficiently compressing high-dimensional video data into semantic tokens that are comprehensible to LLMs. Compared to prior methods which resort to downsampling the original video or aggregating visual tokens using resamplers, leading to information loss and entangled semantics, VideoOrion not only offers a more natural and efficient way to derive compact, disentangled semantic representations but also enables explicit object modeling of video content with minimal computational cost. Moreover, the introduced object tokens naturally allow VideoOrion to accomplish video-based referring tasks. Experimental results show that VideoOrion can learn to make good use of the object tokens, and achieves competitive results on both general video question answering and video-based referring benchmarks.

cs.CV↗

arXiv2025

From Pixels to Tokens: Byte-Pair Encoding on Quantized Visual Modalities

Multimodal Large Language Models have made significant strides in integrating visual and textual information, yet they often struggle with effectively aligning these modalities. We introduce a novel image tokenizer that bridges this gap by applying the principle of Byte-Pair Encoding (BPE) to visual data. Unlike conventional approaches that rely on separate visual encoders, our method directly incorporates structural prior information into image tokens, mirroring the successful tokenization strategies used in text-only Large Language Models. This innovative approach enables Transformer models to more effectively learn and reason across modalities. Through theoretical analysis and extensive experiments, we demonstrate that our BPE Image Tokenizer significantly enhances MLLMs' multimodal understanding capabilities, even with limited training data. Leveraging this method, we develop Being-VL-0, a model that demonstrates superior performance across various benchmarks and shows promising scalability, potentially paving the way for more efficient and capable multimodal foundation models.

cs.AI↗

arXiv2025

Influence of the chiral magnetic effect on particle-pair elliptic anisotropy

Chiral Magnetic Effect (CME) is a phenomenon in which electric charge is separated by a strong magnetic field from local domains of chirality imbalance in quantum chromodynamics. The CME-sensitive, azimuthal correlator difference $Δγ$ between opposite-sign (OS) and same-sign (SS) charged hadron pairs is contaminated by a major physics background proportional to the particle elliptic anisotropy ($v_2$). The CME signal, on the other hand, contributes to the difference in the pair elliptic anisotropies between OS and SS pairs ($Δv_{2,\rm pair}$). We investigate $Δv_{2,\rm pair}$ and found its sensitivity to CME to be similar to that of the $Δγ$ observable.

hep-ph↗

arXiv2025

Azimuthal Correlation Anisotropies in p + p Collisions Simulated by Pythia

Stimulated by the keen interest of possible collective behavior in high-energy proton-proton and proton-nucleus collisions, we study two-particle angular correlations in pseudorapidity and azimuthal differences in simulated p + p interactions by the Pythia 8 event generator. Multi-parton interactions and color connection are included in these simulations which have been perceived to produce collectivity in final-state particles. Meanwhile, contributions from genuine few-body nonflow correlations, not of collective flow behavior, are known to be severe in those small-system collisions. We present our Pythia correlation studies in a pedagogical way and report azimuthal harmonic anisotropies analyzed by several methods. We observe anisotropies in those Pythia simulated events qualitatively and semi-quantitatively similar to experimental data. Our findings highlight the delicate nature of azimuthal anisotropies in small-system collisions, and provide a benchmark helping improve data analysis and interpret experimental measurements in small-system collisions.

nucl-th↗

arXiv2025

Impact of Tracking Resolutions on $ϕ$-Meson Spin Alignment Measurement

Measurements of global spin alignment of vector mesons in relativistic heavy-ion collisions can provide unique insights into spin-orbit interactions and vector meson dynamics in the Quark-Gluon Plasma (QGP) produced in those collisions. The global spin alignment is measured by the $00^{\rm th}$ coefficient of the spin density matrix, $ρ_{00}$, via the polar angle ($θ^{*}$) of the decay-daughter momentum in the parent rest frame with respect to the direction of the orbital angular momentum of the collision. Such measurements are affected by the angular and momentum resolutions of the reconstructed tracks in the experiment. Such effects are nontrivial because of kinematic complications caused by the boost to the parent rest frame, and could be important given that the global spin alignment signal is weak. In this paper, we investigate the effects of experimental tracking resolutions on measurements of the $ϕ$(1020) meson $ρ_{00}$. We study these effects for two methods of $ρ_{00}$ measurements, the conventional method analyzing the $ϕ$-meson yield versus $\cos^2 θ^*$ and the invariant mass ($m_{\rm inv}$) method utilizing $\langle\cos^2θ^*\rangle$ versus $m_{\rm inv}$. Using typical resolution values from experiments, we find that the effect of track resolution on $ρ_{00}$ is small, well within typical measurement uncertainties.

physics.data-an↗

arXiv2024

Echo: Simulating Distributed Training At Scale

Simulation offers unique values for both enumeration and extrapolation purposes, and is becoming increasingly important for managing the massive machine learning (ML) clusters and large-scale distributed training jobs. In this paper, we build Echo to tackle three key challenges in large-scale training simulation: (1) tracing the runtime training workloads at each device in an ex-situ fashion so we can use a single device to obtain the actual execution graphs of 1K-GPU training, (2) accurately estimating the collective communication without high overheads of discrete-event based network simulation, and (3) accounting for the interference-induced computation slowdown from overlapping communication and computation kernels on the same device. Echo delivers on average 8% error in training step -- roughly 3x lower than state-of-the-art simulators -- for GPT-175B on a 96-GPU H800 cluster with 3D parallelism on Megatron-LM under 2 minutes.

cs.LG↗

arXiv2024

Review of nonflow estimation methods and uncertainties in relativistic heavy-ion collisions

Collective anisotropic flow, where particles are correlated over the entire event, is a prominent phenomenon in relativistic heavy-ion collisions and is sensitive to the properties of the matter created in those collisions. It is often measured by two- and multi-particle correlations and is therefore contaminated by nonflow, those genuine few-body correlations unrelated to the global event-wise correlations. Many methods have been devised to estimate nonflow contamination with various degrees of successes and difficulties. Here, we review those methods pedagogically, discussing the pros and cons of each method, and give examples of ballpark estimate of nonflow contamination and associated uncertainties in relativistic heavy-ion collisions. We hope such a review of the various nonflow estimation methods in a single place would prove helpful to future researches.

nucl-ex↗

arXiv2024

Catalan Numbers, Riccati Equations and Convergence

We analyze both finite and infinite systems of Riccati equations derived from stochastic differential games on infinite networks. We discuss a connection to the Catalan numbers and the convergence of the Catalan functions by Fourier transforms.

math.CA↗

arXiv2024

UniCode: Learning a Unified Codebook for Multimodal Large Language Models

In this paper, we propose \textbf{UniCode}, a novel approach within the domain of multimodal large language models (MLLMs) that learns a unified codebook to efficiently tokenize visual, text, and potentially other types of signals. This innovation addresses a critical limitation in existing MLLMs: their reliance on a text-only codebook, which restricts MLLM's ability to generate images and texts in a multimodal context. Towards this end, we propose a language-driven iterative training paradigm, coupled with an in-context pre-training task we term ``image decompression'', enabling our model to interpret compressed visual data and generate high-quality images.The unified codebook empowers our model to extend visual instruction tuning to non-linguistic generation tasks. Moreover, UniCode is adaptable to diverse stacked quantization approaches in order to compress visual signals into a more compact token representation. Despite using significantly fewer parameters and less data during training, Unicode demonstrates promising capabilities in visual reconstruction and generation. It also achieves performances comparable to leading MLLMs across a spectrum of VQA benchmarks.

cs.CV↗

arXiv2024

Azimuthal distribution of exponential format for particle collective motions in heavy-ion collisions under asynchronous assumption

Particle azimuth distributions are widely studied in heavy-ion collisions. They are often expanded in Fourier series to extract anisotropic flow harmonics simultaneously. It was recently proposed that the different orders of flows could happen asynchronously or noninterdependently. This study extends this idea to an exponential format of the azimuth distribution, which makes it straightforward to extract the asynchronous flow coefficients. We compare these new coefficients to the conventional ones, and find consistency in the leading coefficient and discrepancy in higher-order ones.

nucl-th↗

arXiv2023

Steve-Eye: Equipping LLM-based Embodied Agents with Visual Perception in Open Worlds

Recent studies have presented compelling evidence that large language models (LLMs) can equip embodied agents with the self-driven capability to interact with the world, which marks an initial step toward versatile robotics. However, these efforts tend to overlook the visual richness of open worlds, rendering the entire interactive process akin to "a blindfolded text-based game." Consequently, LLM-based agents frequently encounter challenges in intuitively comprehending their surroundings and producing responses that are easy to understand. In this paper, we propose Steve-Eye, an end-to-end trained large multimodal model designed to address this limitation. Steve-Eye integrates the LLM with a visual encoder which enables it to process visual-text inputs and generate multimodal feedback. In addition, we use a semi-automatic strategy to collect an extensive dataset comprising 850K open-world instruction pairs, empowering our model to encompass three essential functions for an agent: multimodal perception, foundational knowledge base, and skill prediction and planning. Lastly, we develop three open-world evaluation benchmarks, then carry out extensive experiments from a wide range of perspectives to validate our model's capability to strategically act and plan. Codes and datasets will be released.

cs.CV↗

arXiv2023

LLaMA Rider: Spurring Large Language Models to Explore the Open World

Recently, various studies have leveraged Large Language Models (LLMs) to help decision-making and planning in environments, and try to align the LLMs' knowledge with the world conditions. Nonetheless, the capacity of LLMs to continuously acquire environmental knowledge and adapt in an open world remains uncertain. In this paper, we propose an approach to spur LLMs to explore the open world, gather experiences, and learn to improve their task-solving capabilities. In this approach, a multi-round feedback-revision mechanism is utilized to encourage LLMs to actively select appropriate revision actions guided by feedback information from the environment. This facilitates exploration and enhances the model's performance. Besides, we integrate sub-task relabeling to assist LLMs in maintaining consistency in sub-task planning and help the model learn the combinatorial nature between tasks, enabling it to complete a wider range of tasks through training based on the acquired exploration experiences. By evaluation in Minecraft, an open-ended sandbox world, we demonstrate that our approach LLaMA-Rider enhances the efficiency of the LLM in exploring the environment, and effectively improves the LLM's ability to accomplish more tasks through fine-tuning with merely 1.3k instances of collected data, showing minimal training costs compared to the baseline using reinforcement learning.

cs.LG↗

arXiv2023

Learning Multi-Object Positional Relationships via Emergent Communication

The study of emergent communication has been dedicated to interactive artificial intelligence. While existing work focuses on communication about single objects or complex image scenes, we argue that communicating relationships between multiple objects is important in more realistic tasks, but understudied. In this paper, we try to fill this gap and focus on emergent communication about positional relationships between two objects. We train agents in the referential game where observations contain two objects, and find that generalization is the major problem when the positional relationship is involved. The key factor affecting the generalization ability of the emergent language is the input variation between Speaker and Listener, which is realized by a random image generator in our work. Further, we find that the learned language can generalize well in a new multi-step MDP task where the positional relationship describes the goal, and performs better than raw-pixel images as well as pre-trained image features, verifying the strong generalization ability of discrete sequences. We also show that language transfer from the referential game performs better in the new task than learning language directly in this task, implying the potential benefits of pre-training in referential games. All in all, our experiments demonstrate the viability and merit of having agents learn to communicate positional relationships between multiple objects through emergent communication.

cs.LG↗