GigaChat Family: Efficient Russian Language Modeling Through Mixture of Experts Architecture

Generative large language models (LLMs) have become crucial for modern NLP research and applications across various languages. However, the development of foundational models specifically tailored to the Russian language has been limited, primarily due to the significant computational resources required. This paper introduces the GigaChat family of Russian LLMs, available in various sizes, including base models and instruction-tuned versions. We provide a detailed report on the model architecture, pre-training process, and experiments to guide design choices. In addition, we evaluate their performance on Russian and English benchmarks and compare GigaChat with multilingual analogs. The paper presents a system demonstration of the top-performing models accessible via an API, a Telegram bot, and a Web interface. Furthermore, we have released three open GigaChat models in open-source (https://huggingface.co/ai-sage), aiming to expand NLP research opportunities and support the development of industrial solutions for the Russian language.

Solving Data Quality Problems with Desbordante: a Demo

Data profiling is an essential process in modern data-driven industries. One of its critical components is the discovery and validation of complex statistics, including functional dependencies, data constraints, association rules, and others. However, most existing data profiling systems that focus on complex statistics do not provide proper integration with the tools used by contemporary data scientists. This creates a significant barrier to the adoption of these tools in the industry. Moreover, existing systems were not created with industrial-grade workloads in mind. Finally, they do not aim to provide descriptive explanations, i.e. why a given pattern is not found. It is a significant issue as it is essential to understand the underlying reasons for a specific pattern's absence to make informed decisions based on the data. Because of that, these patterns are effectively rest in thin air: their application scope is rather limited, they are rarely used by the broader public. At the same time, as we are going to demonstrate in this presentation, complex statistics can be efficiently used to solve many classic data quality problems. Desbordante is an open-source data profiler that aims to close this gap. It is built with emphasis on industrial application: it is efficient, scalable, resilient to crashes, and provides explanations. Furthermore, it provides seamless Python integration by offloading various costly operations to the C++ core, not only mining. In this demonstration, we show several scenarios that allow end users to solve different data quality problems. Namely, we showcase typo detection, data deduplication, and data anomaly detection scenarios.

Desbordante: from benchmarking suite to high-performance science-intensive data profiler (preprint)

Pioneering data profiling systems such as Metanome and OpenClean brought public attention to science-intensive data profiling. This type of profiling aims to extract complex patterns (primitives) such as functional dependencies, data constraints, association rules, and others. However, these tools are research prototypes rather than production-ready systems. The following work presents Desbordante - a high-performance science-intensive data profiler with open source code. Unlike similar systems, it is built with emphasis on industrial application in a multi-user environment. It is efficient, resilient to crashes, and scalable. Its efficiency is ensured by implementing discovery algorithms in C++, resilience is achieved by extensive use of containerization, and scalability is based on replication of containers. Desbordante aims to open industrial-grade primitive discovery to a broader public, focusing on domain experts who are not IT professionals. Aside from the discovery of various primitives, Desbordante offers primitive validation, which not only reports whether a given instance of primitive holds or not, but also points out what prevents it from holding via the use of special screens. Next, Desbordante supports pipelines - ready-to-use functionality implemented using the discovered primitives, for example, typo detection. We provide built-in pipelines, and the users can construct their own via provided Python bindings. Unlike other profilers, Desbordante works not only with tabular data, but with graph and transactional data as well. In this paper, we present Desbordante, the vision behind it and its use-cases. To provide a more in-depth perspective, we discuss its current state, architecture, and design decisions it is built on. Additionally, we outline our future plans.