The paper presents the tasks to be solved by promising RTC VN in the interests of Rvi. The conclusion is formulated that the problem of training and improving the quality of knowledge of Rvi specialists using military robotic systems remains one of the urgent problems of higher mili-tary professional education and acquires new aspects of consideration. It is shown that one of the effective ways to solve the problem of training and improving the quality of knowledge of Rvi spe-cialists is to develop and implement computer-based training systems with elements of virtual real-ity and 3D visualization of the studied samples of equipment and weapons in the educational pro-cess. The main features of the computer information and reference system "compendium of the Rvi" developed at the Mikhailovsky military artillery Academy and used in the educational process are briefly described. Preliminary results of the conducted pedagogical experiment with the use of the "Rvi compendium" are presented, and the main factors that increase the effectiveness of the educational process are Noted. Based on the results of the pedagogical experiment made a rea-sonable inference that the use of KISS "compendium Rvia" allows to increase learning efficiency, to reduce terms of development of technology that is more efficient use of training time and as a result reduce the cost of training and the number of vehicles.
This paper focuses on overcoming the systemic limitations of the large-scale language model (LLM) scaling paradigm, which are related to data exhaustion and exponential growth in computational costs. This enables the development of more efficient approaches to building NLP models without sacrificing their performance. The goal of this study is to compare the performance of a standard transformer architecture (nanoGPT) and a model using semantic embeddings (nanoSonar) for language modeling tasks under resource constraints. Working with conceptual embeddings allows us to identify deeper linguistic patterns and reduce the amount of required training data, significantly improving modeling efficiency. The study utilized the TinyStories dataset, which includes short narratives with a clear structure. Before implementing the models, the data was preprocessed: for nanoGPT, tokenization was performed using the BPE method, and for nanoSonar, text was converted into semantic embeddings using a pretrained Sonar model. The models were evaluated using the loss and perplexity metrics. The results showed that the nanoSonar model provides significantly lower perplexity (6.609 versus 39.151 for nanoGPT) and demonstrates more robust training dynamics at later stages. This paper presents an analysis of modern approaches to scaling optimization (MoE, distillation, PEFT) and promising architectures (LRM, SSM, RWKV), and provides practical recommendations for applying models operating in the space of semantic embeddings to domain-specific problems and systems with limited computational resources. The results of this study can be useful in developing efficient language models that combine high generation quality with a cost-effective architecture.
The relevance of this research stems from the contradiction between the need to increase food production in an urbanized environment and the fragmentation of existing high-tech solutions (hydroponics, aeroponics, IoT), which are being implemented in isolation, without a unified management methodology. The lack of unified approaches to data collection and adaptive control of environmental parameters limits the scalability of vertical farms. The goal of this research is to develop and theoretically substantiate the architecture of a universal adaptive management model for closed-loop agricultural production systems, integrating various cultivation methods based on machine learning algorithms. The methodology is based on a systematic analysis of scientific publications and experimental data on the use of embedded devices and machine learning algorithms in hydroponic, aeroponic, and soil-based vertical greenhouses. Based on this data synthesis, parametric matrices were constructed to standardize technological processes. The main results include the development of a structural diagram of a universal model that enables the integration of disparate systems into a single platform with the ability to continuously monitor and perform predictive analytics. The minimum required sensor set is substantiated: pH, EC, temperature, humidity, CO₂, PAR, pressure, and nutrient solution flow. The proposed architecture enables dynamic switching between hydroponic, aeroponic, and indoor modes within a single phytotron. The conclusions and significance of this work lie in creating a foundation for designing scalable vertical farms with predictable profitability and resource efficiency indicators, while enabling continuous further training of AI algorithms for predictive microclimate management and early plant disease detection in urban environments
