Parallel Combination Research Articles

TPAFNet: Transformer-Driven Pyramid Attention Fusion Network for 3D Medical Image Segmentation.

The field of 3D medical image segmentation is witnessing a growing trend in the utilization of combined networks that integrate convolutional neural networks and transformers. Nevertheless, prevailing hybrid networks are confronted with limitations in their straightforward serial or parallel combination methods and lack an effective mechanism to fuse channel and spatial feature attention. To address these limitations, we present a robust multi-scale 3D medical image segmentation network, the Transformer-Driven Pyramid Attention Fusion Network, which is denoted as TPAFNet, leveraging a hybrid structure of CNN and transformer. Within this framework, we exploit the characteristics of atrous convolution to extract multi-scale information effectively, thereby enhancing the encoding results of the transformer. Furthermore, we introduce the TPAF block in the encoder to seamlessly fuse channel and spatial feature attention from multi-scale feature inputs. In contrast to conventional skip connections that simply concatenate or add features, our decoder is enriched with a TPAF connection, elevating the integration of feature attention between low-level and high-level features. Additionally, we propose a low-level encoding shortcut from the original input to the decoder output, preserving more original image features and contributing to enhanced results. Finally, the deep supervision is implemented using a novel CNN-based voxel-wise classifier to facilitate better network convergence. Experimental results demonstrate that TPAFNet significantly outperforms other state-of-the-art networks on two public datasets, indicating that our research can effectively improve the accuracy of medical image segmentation, thereby assisting doctors in making more precise diagnoses.

Layered V10O24·nH2O as a new ammonium ion host material for aqueous and quasi-solid-state ammonium-ion hybrid capacitors

Aqueous ammonium-ion hybrid capacitors (AIHCs) emerge as a promising candidate for efficient and sustainable charge storage. Yet, a pivotal challenge lies in the development of an ammonium ion (NH4+) host material that possesses high capacity and prolonged cycle life. Herein, we explore the potential of layered V10O24·nH2O, synthesized via electrodeposition method, for use in high-capacity aqueous NH4+ storage. Our investigation, supported by both experimental observations and first-principles theoretical computations, has demonstrated that the layered V10O24·nH2O exhibits commendable electrochemical performance for aqueous NH4+ storage, which is attributed to the presence of certain amount of crystal water and unique nanowire network structure. This allows for a high specific capacity of 203.1 mAh g−1 at 0.3 A g−1, and the capacity retention is 97.6 % after 20,000 cycles. Ex-situ characterizations uncovered that the insertion and extraction of NH4+ in the V10O24·nH2O layers are reversible processes, accompanied by changes in the oxidation states of vanadium and the reversible contraction and expansion of the interlayer spacing. Furthermore, the as-assembled AIHCs device not only exhibits a voltage window of 1.8 V, surpassing other types of hybrid capacitors, but also retains 82.8 % capacitance after 10,000 cycles, showcasing its good flexibility and potential for series–parallel combinations. The study highlights the potential of layered V10O24·nH2O electrodes for sustainable energy storage applications and offers some insights for high-performance hybrid capacitors with non-metallic NH4+ as the charge carrier.

Evaluation of reliability and characteristics measure using exponential decay distribution of non‐series parallel network

AbstractThe purpose of this paper is to analyze the reliability and characteristics of the non‐series‐parallel network. The illustrative non‐series parallel networks, that is, the Twin T‐Network and the network with six components of resistor and capacitor, have been converted into a series of parallel network combinations by use of the minimal path technique. The stochastic variables related to component failure rate have been demonstrated using exponential decay distribution. We obtain the reliability, mean time to failure, and sensitivity with the use of the exponential decay distribution. The reliability and its attributes are examined by assuming different values of scale and shape parameters at different points of time. Using Owen's methodology, we obtained minimal signature, tail signature, expected time, expected cost, and Barlow‐Proschan index were determined. The performance of all reliability characteristics has been observed graphically for arbitrary values of parameters. A real‐world example is used to introduce the point.

Detection of Protein Markers From Blood Samples of Cervical Cancer Patients.

Objective Human papillomavirus (HPV) is the prevalent cause of cervical cancer in females worldwide, necessitating the development of fast and reliable diagnostic methods for early detection of HPV. The study aims to detect serum proteins like squamous cell carcinoma antigen (SCCA/SerpinB3), cytokeratin fragment antigen 21-1 (CYFRA 21-1), carcinoembryonic antigen (CEA), and high mobility group box chromosomal protein 1 (HMGB1) as biomarkers and their combination concerning the type of HPV. Methods Serum samples from a total of 36 cervical cancer patients were initially subjected to sodium dodecyl sulfate-polyacrylamide gel electrophoresis (SDS-PAGE), followed by western blotting with anti-Serpin B3/SCCA, anti-CEA, anti-HMGB, and anti-CYFRA 21-1.The frequency of samples for each protein was obtained. In the Chi-square test, correlations with p-values less than 0.05 were deemed statistically significant. Results Irrespective of HPV type, the CEA had the highest percentage of definite responses (58.33%). Subsequently, SCCA, HMGB1, and Cytokeratin-19 protein presented 47.22%, 27.78%, and 25% responses, respectively. HPV types influenced thedifference inprotein combinations, with HPV-16 presenting the most positive responses followed by HPV-(16+18). HPV-18 presented the least number of affirmative responses. Conclusion Our study presents the CEA protein as a possible biomarker and HPV-16 as the most prominent HPV type to different parallel combinations of serum proteins. The protein combinations canbe appliedto future cancer detection and therapy.

Experimental investigation on the low temperature performance of a novel three-pressure air source heat pump

As people’s demand for living comfort grows, the problems of conventional air source heat pumps (ASHPs) become more prominent in low-temperature and high-humidity environments: heating performance attenuation, low defrosting efficiency, and significant impact on indoor thermal comfort during defrosting. Existing literature usually focuses on only one or two of these issues, and few comprehensively cover them. Based on current literature, this paper proposes a novel multi-functional three-pressure ASHP that can cover the above three aspects with a series–parallel combination structure featuring dual compressors and dual cycles. This design enables switching multiple heating and defrosting modes through valve combination and improves performance through two-stage throttling and subcooling. An experimental platform is constructed to evaluate its operational performance. The experimental results show that: (1) In heating modes, the average heating capacity of the three-pressure cycle heating mode is 2897.15 W, and the average COP is 2.47, while these two data of the conventional cycle heating mode are 2393.65 W and 2.4, and the improvement rates of 20.4 % and 2.9 %, respectively. It also extends the high-efficiency heating period of the system, enables rapid heating at start-up, increases compressor suction and discharge pressure, and improvs system stability. (2) In defrosting modes, compared with the conventional defrosting cycle, the three-pressure defrosting cycle raises the condensation temperature faster, reduces the defrosting time from 448 s to 265 s (40.8 % shorter). The auxiliary defrosting cycle has little effect on indoor thermal comfort and less defrosting power consumption, with the power saving rate as high as 44.6 %. Overall, the novel three-pressure ASHP has a unique and innovative structure, significantly optimizing performance under low-temperature and frosting conditions and improving energy efficiency and comfort. The results of this study provide a new solution concept for the future development of ASHPs, offering significant reference value for further development and application.

Local and global mixture network for image inpainting

In general, CNN-based inpainting can recover local patterns effectively using convolutional filters, but it may not exploit global correlation fully. On the other hand, transformer-based inpainting can fill in large holes faithfully based on global correlation, rather than local one. In this paper, we propose a novel image inpainting algorithm, called local and global mixture (LGM), to take advantage of the strengths of both approaches and compensate for their weaknesses. The LGM network comprises the local inpainting network (LIN) and the global inpainting network (GIN) in parallel, which are based on convolutional layers and transformer blocks, respectively, and exchange their intermediate results with each other. Furthermore, we develop an error propagation model with a continuous error mask, updated in LIN but used in both LIN and GIN to provide more reliable inpainting results. Extensive experiments demonstrate that the proposed LGM algorithm provides excellent inpainting performance, which indicates the efficacy of the parallel combination of LIN and GIN and the effectiveness of the error propagation model.

Adsorption of phenolic compounds on polymeric ionic liquids: Adsorption isotherms interpretation and thermodynamic study

This study describes the thermodynamic assessment and steric interpretation of the 4-nitrophenol (4-NP) and 4-chlorophenol (4-CP) adsorption on polymeric ionic liquids (PILs) using advanced statistical physics model. A monolayer model with a single energy level was employed to elucidate the adsorption mechanisms of these phenolic compounds. The results showed that the orientation of 4-NP and 4-CP on the surface of this adsorbent was a combination of parallel and non-parallel configurations at the tested adsorption temperatures. The values of the adsorption capacities at saturation were 484.7 and 406.7 mg/g for4-NP and 4-CP, respectively. The modelling results indicated that more active sites from PILs surface were involved in the removal of 4-nitrophenol than those required for the 4-chlorophenol adsorption. The calculation of adsorption energies confirmed that the adsorption of 4-NP and 4-CP on PILs was exothermic and driven by physical forces involving hydrogen bonds and van der Waals forces. Three thermodynamic functions were analyzed to complete the macroscopic analysis of the adsorption mechanism for these relevant phenolic contaminants.

Marking-Based Perpendicular Parking Slot Detection Algorithm Using LiDAR Sensors

The emergence of automotive-grade LiDARs has given rise to new potential methods to develop novel advanced driver assistance systems (ADAS). However, accurate and reliable parking slot detection (PSD) remains a challenge, especially in the low-light conditions typical of indoor car parks. Existing camera-based approaches struggle with these conditions and require sensor fusion to determine parking slot occupancy. This paper proposes a parking slot detection (PSD) algorithm which utilizes the intensity of a LiDAR point cloud to detect the markings of perpendicular parking slots. LiDAR-based approaches offer robustness in low-light environments and can directly determine occupancy status using 3D information. The proposed PSD algorithm first segments the ground plane from the LiDAR point cloud and detects the main axis along the driving direction using a random sample consensus algorithm (RANSAC). The remaining ground point cloud is filtered by a dynamic Otsu’s threshold, and the markings of parking slots are detected in multiple windows along the driving direction separately. Hypotheses of parking slots are generated between the markings, which are cross-checked with a non-ground point cloud to determine the occupancy status. Test results showed that the proposed algorithm is robust in detecting perpendicular parking slots in well-marked car parks with high precision, low width error, and low variance. The proposed algorithm is designed in such a way that future adoption for parallel parking slots and combination with free-space-based detection approaches is possible. This solution addresses the limitations of camera-based systems and enhances PSD accuracy and reliability in challenging lighting conditions.

Remaining useful life prediction of rolling bearings based on time convolutional network and transformer in parallel

Abstract Remaining useful life prediction (RUL) is crucial for maintaining the reliability and safety of industrial equipment. Recently, the Transformer has been widely used due to its ability to effectively extract global feature information in rolling bearing RUL prediction. However, the Transformer is weak in acquiring local feature information and cannot extract temporal features from the degradation process. Conversely, a temporal convolutional network (TCN) can effectively extract local features but is weak in global feature extraction. Therefore, to address the above problems, this paper proposes a prediction method based on the parallel combination of TCN and Transformer. The method first extracts the time domain, frequency domain, and time-frequency domain features from the original vibration signals. After screening through the evaluation indices, the features are fused using K-means clustering and principal component analysis (PCA) to construct health indicators (HI) that characterize the degradation of rolling bearings. Then, a TCN-Transformer parallel network model is constructed to extract both local and global features, and a feed-forward neural network (FNN) is used for the prediction of RUL. Finally, experimental validation is carried out on the PHM2012 bearing dataset, and the results show that the method achieves higher RUL prediction accuracy compared to other methods.

System-Level Implementation of a Parallel-Path Hybrid Switched-Capacitor Amplifier with an Embedded Successive Approximation Register for IoT Applications

A system-level implementation of a parallel-path hybrid switched-capacitor amplifier is presented in this paper. The proposed parallel-path amplifier incorporates a gain and slew rate-boosting switching path in parallel with an embedded assisted SAR path, aiming for IoT applications. As an alternative concept to the conventional analog topologies, the proposed amplifier combines nonlinear and linear paths to provide coarse and fine amplifications. In the coarse amplification, a high current is provided through a switching path for a fraction of time, which improves the slew rate and open-loop DC gain without adding significant static current. Moreover, high accuracy is achieved through the embedded assisted SAR path, which provides a resolution of 1/2N. In addition, each extra bit of the embedded SAR path improves the total open-loop DC gain by 6 dB. The theory of operation is performed to study how the switching and assisted SAR paths can enhance the amplifier’s settling error. In addition, an existence trade-off between the coarse amplification error and the capacitive digital-to-analog converter’s number of bits is investigated. The theory and system-level simulation show that the gain and slewing restrictions of the conventional topologies, especially in advanced CMOS technology, can be handled much easier by this parallel combination, where the switching path and assisted SAR path combination provides a high slewing capability and high DC open-loop gain.

Performance Analysis of Multiple Energy-Storage Devices Used in Electric Vehicles

Considering environmental concerns, electric vehicles (EVs) are gaining popularity over conventional internal combustion (IC) engine-based vehicles. Hybrid energy-storage systems (HESSs), comprising a combination of batteries and supercapacitors (SCs), are increasingly utilized in EVs. Such HESS-equipped EVs typically outperform standard electric vehicles. However, the effective management of power sources to meet varying power demands remains a major challenge in the hybrid electric vehicles. This study presents the development of a MATLAB Simulink model for a hybrid energy-storage system aimed at alleviating the load on batteries during periods of high power demand. Two parallel combinations are investigated: one integrating the battery with a supercapacitor and the other with a photovoltaic (PV) system. These configurations address challenges encountered in EVs, such as power fluctuations and battery longevity issues. Although batteries are commonly used in conjunction with solar PV systems for energy storage, they incur higher operating costs due to the necessity of converters. The findings suggest that the proposed supercapacitor–battery configuration reduces battery peak power consumption by up to 39%. Consequently, the supercapacitor–battery HESS emerges as a superior option, possibly prolonging battery cycle life by mitigating stress induced by fluctuating power exchanges during the charging and discharging phases.

A Cooperative Intrusion Detection System for the Internet of Things Using Convolutional Neural Networks and Black Hole Optimization.

Maintaining security in communication networks has long been a major concern. This issue has become increasingly crucial due to the emergence of new communication architectures like the Internet of Things (IoT) and the advancement and complexity of infiltration techniques. For usage in networks based on the Internet of Things, previous intrusion detection systems (IDSs), which often use a centralized design to identify threats, are now ineffective. For the resolution of these issues, this study presents a novel and cooperative approach to IoT intrusion detection that may be useful in resolving certain current security issues. The suggested approach chooses the most important attributes that best describe the communication between objects by using Black Hole Optimization (BHO). Additionally, a novel method for describing the network's matrix-based communication properties is put forward. The inputs of the suggested intrusion detection model consist of these two feature sets. The suggested technique splits the network into a number of subnets using the software-defined network (SDN). Monitoring of each subnet is done by a controller node, which uses a parallel combination of convolutional neural networks (PCNN) to determine the presence of security threats in the traffic passing through its subnet. The proposed method also uses the majority voting approach for the cooperation of controller nodes in order to more accurately detect attacks. The findings demonstrate that, in comparison to the prior approaches, the suggested cooperative strategy can detect assaults in the NSLKDD and NSW-NB15 datasets with an accuracy of 99.89 and 97.72 percent, respectively. This is a minimum 0.6 percent improvement.

Proposal of a scissor-based model for the non-linear analysis of RC beam-column joints strengthened by FRP

The paper presents a new simple modeling approach for studying the monotonic response of RC beam-column joints externally strengthened by Fiber Reinforced Polymer (FRP) materials. The approach assumes a parallel combination of the behavior of the joint in the unstrengthened configuration and the contribution of strengthening system. To achieve this, the common scissor model, where the behavior of joint shear panel is modeled through a rotational spring, is here modified by introducing an additional spring, arranged parallel to the concrete spring, to account for the contribution of strengthening system. Regarding the behavior of the additional spring, the authors proposed a multilinear simplified constitutive law by appropriately combining analytical models available from the literature with specific guidelines provided by Italian technical standards. The proposed model, implemented in the computer code OpenSees and validated against experimental case studies from the literature, demonstrates its strong capability to capture the contribution of the strengthening system on both the strength and ductility of joints.

RLC resonator with diode nonlinearity: Bifurcation comparison of numerical predictions and circuit measurements

A nonlinear RLC resonator is investigated experimentally and numerically using bifurcation analysis. The nonlinearity is due to the parallel combination of a semiconductor rectifier diode and a fixed capacitor. The diode’s junction capacitance, diffusion capacitance, and DC current–voltage relation each contribute to the nonlinearity. The closely related RL-diode resonator has been of interest for many years since its demonstration of period-doubling cascades to chaos. In this study, a direct comparison is made of dynamical regime maps produced from simulations and circuit measurements. The maps show the variety of limit cycles, their bifurcations, and regions of chaos over the 2D parameter space of the source voltage’s frequency and amplitude. The similar structures of the simulated and experimental maps suggest that the diode models commonly used in circuit simulators (e.g., SPICE) work well in bifurcation analyses, successfully predicting complex and chaotic dynamics detected in the circuit. These results may be useful for applications of varactor-loaded split ring resonators.

A Review of Impedance Spectroscopy Technique: Applications, Modelling, and Case Study of Relative Humidity Sensors Development

Impedance Spectroscopy (IS) is a general term for the technique referring to small-signal measurements of the linear electrical response of a domain of interest. This method is based on the analysis of the system’s electrical response to yield helpful information about its domain-dependent physicochemical properties (generally, the analysis is carried out in the frequency domain). Nowadays, there are many areas of application where IS can be used to evaluate or enhance the development of emerging products and processes. As a contribution to this field of research, this paper presents relevant theoretical–practical aspects of the interpretation and analysis of the electrical behavior of materials based on IS and IS modelling. The work starts by historically introducing IS and then goes through different domains of application of the technique, such as Materials Science and correlated areas. Afterwards, an introduction to IS usage for constructing equivalent electrical circuits is presented, aiming at modelling the materials’ electrical behavior, followed by examples from the literature that use the two possible circuit development approaches, the series and the parallel association of circuit elements. Lastly, the authors present a case study of their most recent efforts of a circuit model development of relative humidity (RH) sensors based on heterogeneous mixed metal oxide (MMO) nanostructures, used to understand and identify existing contributions to the overall electrical response of the sensors to moisture; in their case, the electrical response of the MMO sensors was modelled with a high level of superposition between the experimental and fitted data, using a parallel combination of circuit elements, which is an unconventional one.

Innovative heat management method and metaheuristic algorithm optimized power supply-demand balance for PEMFC-ASHP-CHP system

The evolution of distributed building energy systems fuels the growing demand for sustainable energy solutions. In this paper, Proton Exchange Membrane Fuel Cell (PEMFC) and Air Source Heat Pump (ASHP) were integrated to form PEMFC-ASHP-CHP systems in three combination methods, i.e., Direct Combination (DC), Parallel Combination (PC), Series Combination (SC). And compared the energy management strategies and power balance of the system. To further improve reliability and flexibility, a diversion PEMFC-ASHP-CHP system was proposed by combining PC and SC's system advantages. Additionally, an iterative algorithm addressed the mismatch between power supply and demand. An empirical formula was proposed to improve iterative convergence speed for practical control situations. The coefficients were optimized using six metaheuristic algorithms, and the outcomes were summarized into an optimized operational plane to enhance the system control response speed further. The results show that the PC method performs better than DC and SC. It achieves a power consumption reduction of 52.8% and a COP improvement of 111.4% compared with the ASHP system. Meanwhile, the diversion system can more effectively utilize waste heat from the PEMFC and further improve the system's performance. The Queuing Search Algorithm (QSA) demonstrates superior accuracy for coefficient optimization. By using the established empirical formula, the convergence speeds of global and local iterative methods are improved by 26.10% and 41.78%, respectively, compared to the direct iterative algorithm. Ultimately, the optimized operational plane can achieve a maximum 37.16% hydrogen consumption reduction compared to the unoptimized system.

Flexible urea biosensor based on CuS/Bioglass 45S5.TiO2 Nps thin films

Chronic renal insufficiency has been a high-incidence disease around the world in recent years. With the lack of devices to detect these illnesses opportunely, developing urea biosensors is imperative. This work reports the application of copper sulfide (CuS) thin films deposited via a chemical bath and Bioglass 45S5/titanium dioxide nanoparticles thin films in a urea flexible biosensor. The Bioglass 45S5 was deposited over CuS and applied onto polyethylene naphthalate (PEN) and polyethylene terephthalate (PET) with indium tin oxide (ITO) substrates. The systems PEN/CuS/Bioglass 45S5.TiO2 NpS and PET/ITO/CuS/ Bioglass 45S5.TiO2 were characterized optically by UV-Vis. The CuS presented an absorption border around 600 nm and a decrease in transmittance percentage by adding Bioglass. Scanning Electron Microscopy carried out the microstructural characterization; both systems displayed a porous morphology featuring channels suitable for immobilizing the enzyme. The sensors showed a resistive behavior based on the current-voltage curve. The urease enzyme was immobilized by adsorption. The optical changes after adding urease were measured by UV-Vis, obtaining ammonia from the reaction between urease and urea, showing an absorption border around 850 nm. Changes in urea concentrations between 1 and 10 mM caused an increase in resistivity in both samples, obtaining a sensitivity of 12.887 kΩ/sq/mM and 12.411 kΩ/sq/mM, and a linearity of 0.9946 % and 0.9601 % for the samples deposited over PEN and PET ITO, respectively. A Cole-Cole plot showed a behavior corresponding to an equivalent circuit composed of a parallel combination of a resistor and a capacitor in series with a phase constant element; this latter could be attributed to the measurement electrodes. Each sensor was implemented in a Wheatstone bridge. A voltage decrease was obtained for the sample deposited over PET ITO, with a sensitivity of −102.7 mV/mM, and a voltage increase was obtained in the sample deposited over PEN, with a sensitivity of 30.5 mV/mM.

WITHDRAWN: Robust voltage regulation of DC-DC buck converter with ZIP load via an energy shaping control approach

ZIP loads (the parallel combination of constant impedance loads, constant current loads and constant power loads) exist widely in power system. In order to stabilize buck converter based DC distributed system with ZIP load, an adaptive energy shaping controller (AESC) is devised in this paper. Firstly, based on the assumption that the lumped disturbances are known, a full information controller is designed in the framework of the port Hamiltonian system via energy shaping technique. An obvious merit is that it is PI-type controller, which is easy to implement in practical application. Besides, using mathematical deductive method, an estimation of the domain of attraction is given to ensure the strict stability. Furthermore, to eliminate the influence of parameter perturbations on the system, a disturbance observer is proposed to reconstruct the lumped disturbances and then the estimated terms are introduced to above controller to form an adaptive control scheme. In addition, the strict stability analysis of the closed-loop system is given. Lastly, the simulation and experiment results are presented for assessing the designed controller.

Growth rate effect on the dielectric properties of FeNbO4 fibres processed by the laser floating zone technique

Fe-based materials exhibiting high dielectric constants have been attracting great attention due to their potential for application in electronic devices. In this work, iron niobate (FeNbO4) fibres were produced by the Laser Floating Zone technique, applying three different growth speeds: 5, 10 and 25 mm/h. The XRD patterns showed the presence of two phases, being the FeNbO4 the major one in all the samples. The effect of the growth speed is well perceptible in the morphology of the fibres. The dielectric measurements revealed that all samples have at least one dielectric relaxation phenomenon that is thermally activated, with the activation energy, estimated through the Arrhenius law, presenting values between 0.057 and 0.072 eV. The Nyquist plots showed the presence of a single semicircle for each temperature, that could be modelled by an equivalent electrical circuit consisting of an offset resistance in series with a parallel combination of a grain resistance and a grain constant phase element. The grain resistance increased with the increasing growth speed, from 574 to 64497 Ω, with the constant phase element showing the opposite trend, decreasing from 0.40 to 0.17 nF.

Thorough Characterization and Analysis of Large Transformer Model Training At-Scale

Large transformer models have recently achieved great success across various domains. With a growing number of model parameters, a large transformer model training today typically involves model sharding, data parallelism, and model parallelism. Thus, the throughput of large-scale model training depends heavily on the network bandwidth since a combination of model sharding and multiple parallelism strategies incurs various costs. However, prior characterizations of transformer models on high-bandwidth DGX machines that use TFLOPS as a metric may not reflect the performance of a system with lower bandwidth. Furthermore, data and model parallelism reveal significantly distinct training profiles on different system bandwidths at scale and, thus, need a thorough study. In this paper, we provide a bottom-up breakdown of training throughput into compute and communication time, and quantitatively analyze their respective influences on overall end-to-end training scaling. Our evaluation involves an in-depth exploration of data parallelism, scaling up to 512 GPUs with limited bandwidth, and examines three model sharding strategies among six model sizes. We also evaluate three combinations of model parallelism on both high and low bandwidth supercomputing systems. Overall, our work provides a broader perspective on large-scale transformer model training, and our analysis and evaluation yield practical insights for predicting training scaling, shaping the future development of supercomputing system design.