Search for: All records

Transformer models have been widely investigated in different domains by providing long-range dependency handling and global contextual awareness, driving the development of popular AI applications such as ChatGPT, Gemini, and Alexa. State Space Models (SSMs) have emerged as strong contenders in the field of sequential modeling, challenging the dominance of Transformers. SSMs incorporate a selective mechanism that allows for dynamic parameter adjustment based on input data, enhancing their performance. However, this mechanism also comes with increasing computational complexity and bandwidth demands, posing challenges for deployment on resource-constraint mobile devices. To address these challenges without sacrificing the accuracy of the selective mechanism, we propose a sparse learning framework that integrates architecture-aware compiler optimizations. We introduce an end-to-end solution–C 4 n kernel sparsity, which prunes n elements from every four contiguous weights, and develop a compiler-based acceleration solution to ensure execution efficiency for this sparsity on mobile devices. Based on the kernel sparsity, our framework generates optimized sparse models targeting specific sparsity or latency requirements for various model sizes. We further leverage pruned weights to compensate for the remaining weights, enhancing downstream task performance. For practical hardware acceleration, we propose C 4 n -specific optimizations combined with a layout transformation elimination strategy. This approach mitigates inefficiencies arising from fine-grained pruning in linear layers and improves performance across other operations. Experimental results demonstrate that our method achieves superior task performance compared to other semi-structured pruning methods and achieves up-to 7→ speedup compared to llama.cpp framework on mobile devices. 

Conspectus: The role of quantum mechanical coherences or coherent superposition states in excited state processes has received considerable attention in the last two decades largely due to advancements in ultrafast laser spectroscopy. These coherence effects hold promise for enhancing the efficiency and robustness of functionally relevant processes, even when confronted with strong energy disorder and environmental fluctuations. Understanding coherence deeply drives us to unravel mechanisms and dynamics controlled by order and synchronization at a quantum mechanical level, envisioning optical control of coherence to enhance functions or create new ones in molecular and material systems. In this frontier, the interplay between electronic and vibrational dynamics, specifically the influence of vibrations in directing electronic dynamics, has emerged as the leading principle. Here, two energetically disparate quantum degrees of freedom work in-sync to dictate the trajectory of an excited state reaction. Moreover, with the vibrational degree being directly related to the structural composition of molecular or material systems, new molecular designs could be inspired by tailoring certain structural elements. In the realm of chemical kinetics, our understanding of the dynamics of chemical transformations is underpinned by fundamental theories such as transition state theory, activated rate theory, and Marcus theory. These theories elucidate reaction rates by considering the energy barriers that must be overcome for reactants to transform into products. Those barriers are surmounted by the stochastic nature of energy gap fluctuations within reacting systems, emphasizing that the reaction coordinate—the pathway from reactants to products—is not rigidly defined by a specific vibrational motion but encompasses a diverse array of molecular motions. While less is known about the involvement of specific intramolecular vibrational modes, their significance in certain cases cannot be overlooked. In this Account, we summarize key experimental findings that offer deeper insights into the complex electronic-vibrational trajectories encompassing excited states afforded from state-of-the-art ultrafast laser spectroscopy in three exemplary processes: photo-induced electron transfer, singlet-triplet intersystem crossing, and intramolecular vibrational energy flow in molecular systems. We delve into rapid decoherence—loss of phase and amplitude correlations—of vibrational coherences along promoter vibrations during a sub-picosecond intersystem crossing dynamics in a series of binuclear platinum complexes. This rapid decoherence illustrates the vibration-driven reactive pathways from Franck-Condon state to the curve crossing region. We also explore the generation of new vibrational coherences induced by impulsive reaction dynamics—rather than by the laser pulse—in these systems, which sheds light on specific energy dissipation pathways and thereby on the progression of the reaction trajectory in the vicinity of the curve crossing on the product side. Another property of vibrational coherences, amplitude, reveals how energy can flow from one vibration to another in the electronic excited state of a terpyridine-molybdenum complex hosting a nonreactive dinitrogen substrate. A slight change in vibrational energy triggers a quasi-resonant interaction, leading to constructive wavepacket interference and ultimately intramolecular vibrational redistribution from a Franck-Condon active terpyridine vibration to dinitrogen stretching vibration, energizing the dinitrogen bond. 

Domain experts play an important role in data science, as their knowledge can unlock valuable insights from data. As they often lack technical skills required to analyze data, they need collaborations with technical experts. In these joint efforts, productive collaborations are critical not only in the phase of constructing a data science task, but more importantly, during the execution of a task. This need stems from the inherent complexity of data science, which often involves user-defined functions or machine-learning operations. Consequently, collaborators want various interactions during runtime, such as pausing/resuming the execution, inspecting an operator's state, and modifying an operator's logic. To achieve the goal, in the past few years we have been developing an open-source system called Texera to support collaborative data analytics using GUI-based workflows as cloud services. In this paper, we present a holistic view of several important design principles we followed in the design and implementation of the system. We focus on different methods of sending messages to running workers, how these methods are adopted to support various runtime interactions from users, and their trade-offs on both performance and consistency. These principles enable Texera to provide powerful user interactions during a workflow execution to facilitate efficient collaborations in data analytics. 

Although vibronic coupling phenomena have been recognized in the excited state dynamics of transition metal complexes, their impact on photoinduced electron transfer (PET) remains largely unexplored. This study investigates coherent wavepacket (CWP) dynamics during PET processes in a covalently linked electron donor–acceptor complex featuring a cyclometalated Pt(II) dimer as the donor and naphthalene diimide (NDI) as the acceptors. Upon photoexciting the Pt(II) dimer electron donor, ultrafast broadband transient absorption spectroscopy revealed direct modulation of NDI radical anion formation through certain CWP motions and correlated temporal evolutions of the amplitudes for these CWPs with the NDI radical anion formation. These results provide clear evidence that the CWP motions are the vibronic coherences coupled to the PET reaction coordinates. Normal mode analysis identified that the CWP motions originate from vibrational modes associated with the dihedral angles and bond lengths between the planes of the cyclometalating ligand and the NDI, the key modes altering their p-interaction, consequently influencing PET dynamics. The findings highlight the pivotal role of vibrations in shaping the favorable trajectories for the efficient PET processes.