11563 research outputs found
Sort by
Eka-Eval : a comprehensive evaluation framework for large language models in Indian languages
No full textThe rapid advancement of Large Language Models (LLMs) has intensified the need for evaluation frameworks that address the requirements of linguistically diverse regions, such as India, and go beyond English-centric benchmarks. We introduce EKA-EVAL, a unified evaluation framework that integrates over 35+ benchmarks (including 10 Indic benchmarks) across nine major evaluation categories. The framework provides broader coverage than existing Indian language evaluation tools, offering 11 core capabilities through a modular architecture, seamless integration with Hugging Face and proprietary models, and plug-and-play usability. As the first end-to-end suite for scalable, multilingual LLM benchmarking, the framework combines extensive benchmarks, modular workflows, and dedicated support for low-resource Indian languages to enable inclusive assessment of LLM capabilities across diverse domains. We conducted extensive comparisons against five existing baselines, demonstrating that EKA-EVAL achieves the highest participant ratings in four out of five categories
Textures of the Earth: Notes from rural Birbhum in the short stories of Tarasankar Bandyopadhyay
No full textThe vision of the ‘rural’ in Indian literary and aesthetic traditions is significant for a critical interpretation of literary and cultural texts emerging from India. This article presents a critical reading of select short stories of the Bharatiya Jnanpith Award recipient Tarasankar Bandyopadhyay (1898–1971) in the context of rural writings. Having spent most of his life in Birbhum, Bandyopadhyay internalized the peculiarities of rural Bengal in his stories. Bandyopadhyay’s visualizations of the rural transcend conventional notions, weaving a rich, earthy texture that construct an artistic and symbolic system of language and images drawn from everyday life. The conceptual framework of the present research is supported by the theory of literary carnivalization proposed by the Russian thinker Mikhail Bakhtin (1895–1975). The short stories of Bandyopadhyay might be read in the light of the social realism that weaves elements of ‘grotesqueness’ into the pastorality of rural landscapes. Through close readings of two stories; ‘Tarini Majhi’ (1936) and ‘Santan’ (1938), the study explores certain facets of rural life, highlighting the visceral interplay between survival and crisis. The select stories encapsulate the rich and multifaceted world of Bandyopadhyay’s Birbhum which are at once sensuous, artistic and eccentric, offering profound insights into the lived experiences of its characters. Bakhtin’s concepts of the idyllic and the grotesque in his discussions of Problems of Dostoevsky’s Poetics (1984) and Rabelais and His World (1984) provide the conceptual basis to the present research on Bandyopadhyay’s stories. This study aims to creatively understand aesthetics of the ‘rural’, positioning Bandyopadhyay’s short fiction as a critical space to reimagine the literary representation of rural India and its enduring complexities
Status of G2HDM with right handed neutrino coupling in the light of b→cτνanomalies
No full textRecent experimental measurements of several observables in semileptonic B meson decays have pointed towards the possibility of new physics. The LHCb collaboration has reported a significant deviation, exceeding 3.2 σ , in the combined measurement of the ratio of branching ratios R ( D ) − R ( D ⋆ ) from the predictions of the standard model. Furthermore, other observables, such as R J / ψ , P D ⋆ τ , F D ⋆ L , and R Λ c in the b → c ℓ ν transition, have also exhibited noticeable deviations from the standard model predictions. Motivated by these anomalies in the b → c τ ν transitions, we perform a log-likelihood fit incorporating new physics coming from right-handed neutrino couplings and explored the implications of a charged Higgs boson within a generic two Higgs doublet model (G2HDM). Our comprehensive analysis, focused on the τ ν and b τ ν final states, was performed using the High Luminosity run of the Large Hadron Collider (HL-LHC). We demonstrate that HL-LHC has the sensitivity to exclude the remaining allowed region in G2HDM model in explaining these anomalies with charged Higgs boson coupled to right-handed neutrino
Progressive homogenization and damage modelling of a fibre reinforced composite with a viscoplastic matrix
No full textThe present work proposes a novel approach to perform the progressive computational homogenization and internal damage modelling in a fibre reinforced lamina combined with a separate viscoplastic matrix layer. The proposed formulation is especially useful in the matrix-dominated loading cases, where the viscoplastic strain plays an important role in the computation of homogenized stress field. The layer volume fractions of lamina and pure matrix are first computed, and the macro-scale (homogenized) tangent modulus expression is proposed, incorporating the degradation in lamina and matrix layers, employing Voigt approximation. The applicability of the homogenized tangent modulus is subsequently demonstrated by implementing it in an in-house developed non-linear finite element framework while solving several boundary value problems. It is also demonstrated that, the presented approach can be extended to any layered media having different constitutive responses and inelastic strain
A Novel Principle for Transparent Applications of Force Impulses in Cable-Driven Rehabilitation Systems
No full textA critical requirement for rehabilitation robots is achieving high transparency in user interaction to minimize interference when assistance is unnecessary. Cable-driven systems are a compelling alternative to rigid-link robots due to their lighter weight and reduced inertia, enhancing transparency. However, controlling cable tension forces remains a significant challenge, as these forces directly affect the interaction between the user and the robot. Effective strategies must maintain low tension during non-assistive phases while preventing slackness. This paper introduces PACE-R (Passive Active CablE Robot), a novel lightweight actuation system for cable-driven rehabilitation devices. The PACE-R module utilizes remote actuation and an open-loop, discrete state control, where the cable is coupled to the motor only during active intervention. When not assisting, the cable is passively decoupled from the motor, and a low-stiffness spring maintains minimal tension, enabling high transparency. Benchtop tests showed that the module consistently produced force impulses proportional to motor input with delays not exceeding 15 ms. In the treadmill push-off assistance demonstration, PACE-R contributed about 20% to total ankle moment and power. Transparency analysis revealed negligible interference, with only 1% and 0.5% contributions to peak total ankle moment and power, respectively
Model hubs and beyond: analyzing model popularity, performance, and documentation
No full textWith the massive surge in ML models on platforms like Hugging Face, users often lose track and struggle to choose the best model for their downstream tasks, frequently relying on model popularity indicated by download counts, likes, or recency. We investigate whether this popularity aligns with actual model performance and how the comprehensiveness of model documentation correlates with both popularity and performance. In our study, we evaluated a comprehensive set of 500 Sentiment Analysis models on Hugging Face. This evaluation involved massive annotation efforts, with human annotators completing nearly 80,000 annotations, alongside extensive model training and evaluation. Our findings reveal that model popularity does not necessarily correlate with performance. Additionally, we identify critical inconsistencies in model card reporting: approximately 80% of the models analyzed lack detailed information about the model, training, and evaluation processes. Furthermore, about 88% of model authors overstate their models' performance in the model cards. Based on our findings, we provide a checklist of guidelines for users to choose good models for downstream tasks
Closed Cohen-Macaulay Completion of Binomial Edge Ideals
No full textLet CCM denote the class of closed graphs with Cohen-Macaulay binomial edge ideals and PIG denote the class of proper interval graphs. Then CCM ⊆ PIG. The PIG-completion problem is a classical problem in graph theory as well as in molecular biology, and this problem is known to be NP-hard. In this paper, we study the CCM-completion problem. We give a method to construct all possible CCM-completions of a graph. We find the CCM-completion number and the set of all minimal CCM-completions for a large class of graphs. Moreover, for this class, we give a polynomial-time algorithm to compute the CCM-completion number and a minimum CCM-completion of a given graph. The unmixedness and Cohen-Macaulay properties of binomial edge ideals of induced subgraphs are investigated. Also, we discuss the accessible graph completion and the Cohen-Macaulay property of binomial edge ideals of whisker graphs
Exclusive production of acetone using a copper nanoparticle anchored LSCO perovskite electrocatalyst: cell design and metal-support interaction governed electrocatalysis
No full textThis study explores the enhancement of electrochemical CO2 reduction (CO2ER) using a novel Cu nanoparticle decorated La0.8Sr0.2CoO3 (LSCO) perovskite catalyst. The synthesized Cu/LSCO catalyst exhibits exceptional activity and selectivity for acetone production. A systematic variation in Cu loading revealed a non-linear trend in performance: faradaic efficiency (FE) increased from ∼40% for Cu 10/LSCO to a maximum of ∼93.7% for Cu 20/LSCO but significantly dropped to 7.5% for Cu 30/LSCO. Cu 20/LSCO also delivered a partial current density of −20.28 mA cm−2, making it the most efficient composition. This behavior highlights the importance of optimal Cu loading, where enhanced nanoparticle dispersion and strong metal-support interaction (MSI) result in greater active site availability and improved catalytic performance. In contrast, excessive Cu loading leads to particle agglomeration, and diminished CO2ER activity. Cu 20/LSCO also exhibited stable performance over 40 000 seconds, demonstrating its potential for prolonged CO2 electroreduction and highlighting its viability for sustainable CO2 conversion in renewable energy applications. X-ray absorption spectroscopy (XAS) analysis confirmed the oxidation state and local coordination environment of Cu, providing critical mechanistic insight into the observed performance trend and the role of MSI. The Cu/LSCO catalyst, enhanced by metal-support interaction (MSI) along with the cell geometry, is an effective tool for high FE and liquid product selective electrocatalysis. Utilizing Cu or LSCO alone proves inefficient for CO2ER indicating the role of MSI. This strategy can be a stepping stone for developing electrocatalysts for direct multicarbon products at low overpotentials