Search for: All records

Token-free language models learn directly from raw bytes and remove the inductive bias of subword tokenization. Operating on bytes, however, results in significantly longer sequences. In this setting, standard autoregressive Transformers scale poorly as the effective memory required grows with sequence length. The recent development of the Mamba state space model (SSM) offers an appealing alternative approach with a fixed-sized memory state and efficient decoding. We propose MambaByte, a token-free adaptation of the Mamba SSM trained autoregressively on byte sequences. In terms of modeling, we show MambaByte to be competitive with, and even to outperform, state-of-the-art subword Transformers on language modeling tasks while maintaining the benefits of token-free language models, such as robustness to noise. In terms of efficiency, we develop an adaptation of speculative decoding with tokenized drafting and byte-level verification. This results in a 2.6× inference speedup to the standard MambaByte implementation, showing similar decoding efficiency as the subword Mamba. These findings establish the viability of SSMs in enabling token-free language modeling. 

This paper presents the development of novel rechargeable cement-based batteries with carbon fiber mesh for energy storage applications. With the increasing demand for sustainable energy storage solutions, there is a growing interest in exploring unconventional materials and technologies. The batteries featured the carbon fiber mesh, which coated with nickel oxide and iron materials as electrodes and immersed in a cement-based electrolyte, offering a unique approach to energy storage. Experimental investigations, including electrochemical impedance spectroscopy, cyclic voltammetry, charge-discharge cycling, and rate performance assessments, were conducted to evaluate the batteries' performance. Results indicated that the batteries have promising features such as high ionic conductivity of the cement-based electrolyte and stable charge-discharge behaviors over 100 cycles. Cyclic voltammetry curves demonstrated quasi-reversible redox peaks, indicative of battery-type electrochemistry. The rechargeable cement-based batteries exhibited stability in discharge capacity, efficiency, and energy density, surpassing existing literatures on cement batteries, with a maximum energy density of 7.6 Wh/ m2. Despite challenges related to efficiency and energy density, this paper envisions the practical applications for the batteries, from powering light sensors to supporting 5G base stations and meeting daily electricity needs. Integration of rechargeable cement-based batteries and clean energy sources holds significant promise for global energy storage solutions. In conclusion, this research provides valuable insights into developing rechargeable cement-based batteries, highlights their potential as sustainable energy storage solutions with opportunities for further optimization and future advancements. 

Abstract Biofilms are ubiquitous surface-associated bacterial communities embedded in an extracellular matrix. It is commonly assumed that biofilm cells are glued together by the matrix; however, how the specific biochemistry of matrix components affects the cell-matrix interactions and how these interactions vary during biofilm growth remain unclear. Here, we investigate cell-matrix interactions inVibrio cholerae, the causative agent of cholera. We combine genetics, microscopy, simulations, and biochemical analyses to show thatV. choleraecells are not attracted to the main matrix component (Vibriopolysaccharide, VPS), but can be attached to each other and to the VPS network through surface-associated VPS and crosslinks formed by the protein Bap1. Downregulation of VPS production and surface trimming by the polysaccharide lyase RbmB cause surface remodeling as biofilms age, shifting the nature of cell-matrix interactions from attractive to repulsive and facilitating cell dispersal as aggregated groups. Our results shed light on the dynamics of diverse cell-matrix interactions as drivers of biofilm development. 

Sponges are animals that inhabit many aquatic environments while filtering small particles and ejecting metabolic wastes. They are composed of cells in a bulk extracellular matrix, often with an embedded scaffolding of stiff, siliceous spicules. We hypothesize that the mechanical response of this heterogeneous tissue to hydrodynamic flow influences cell proliferation in a manner that generates the body of a sponge. Towards a more complete picture of the emergence of sponge morphology, we dissected a set of species and subjected discs of living tissue to physiological shear and uniaxial deformations on a rheometer. Various species exhibited rheological properties such as anisotropic elasticity, shear softening and compression stiffening, negative normal stress, and non-monotonic dissipation as a function of both shear strain and frequency. Erect sponges possessed aligned, spicule-reinforced fibres which endowed three times greater stiffness axially compared with orthogonally. By contrast, tissue taken from shorter sponges was more isotropic but time-dependent, suggesting higher flow sensitivity in these compared with erect forms. We explore ecological and physiological implications of our results and speculate about flow-induced mechanical signalling in sponge cells. 

Abstract Cavitation has long been recognized as a crucial predictor, or precursor, to the ultimate failure of various materials, ranging from ductile metals to soft and biological materials. Traditionally, cavitation in solids is defined as an unstable expansion of a void or a defect within a material. The critical applied load needed to trigger this instability -- the critical pressure -- is a lengthscale independent material property and has been predicted by numerous theoretical studies for a breadth of constitutive models. While these studies usually assume that cavitation initiates from defects in the bulk of an otherwise homogeneous medium, an alternative and potentially more ubiquitous scenario can occur if the defects are found at interfaces between two distinct media within the body. Such interfaces are becoming increasingly common in modern materials with the use of multimaterial composites and layer-by-layer additive manufacturing methods. However, a criterion to determine the threshold for interfacial failure, in analogy to the bulk cavitation limit, has yet to be reported. In this work, we fill this gap. Our theoretical model captures a lengthscale independent limit for interfacial cavitation, and is shown to agree with our observations at two distinct lengthscales, via two different experimental systems. To further understand the competition between the two cavitation modes (bulk versus interface), we expand our investigation beyond the elastic response to understand the ensuing unstable propagation of delamination at the interface. A phase diagram summarizes these results, showing regimes in which interfacial failure becomes the dominant mechanism.