Databricks announced the public preview of the Mosaic AI Agent Framework and Agent Evaluation during the Data + AI Summit 2024. These innovative tools aim to assist developers in building and deploying high-quality Agentic and Retrieval Augmented Generation (RAG) applications on the Databricks Data Intelligence Platform. Challenges in Building High-Quality Generative AI Applications Creating a…
The field of language models has seen remarkable progress, driven by transformers and scaling efforts. OpenAI’s GPT series demonstrated the power of increasing parameters and high-quality data. Innovations like Transformer-XL expanded context windows, while models such as Mistral, Falcon, Yi, DeepSeek, DBRX, and Gemini pushed capabilities further. Visual language models (VLMs) have also advanced rapidly.…
Deep learning has demonstrated remarkable success across various scientific fields, showing its potential in numerous applications. These models often come with many parameters requiring extensive computational power for training and testing. Researchers have been exploring various methods to optimize these models, aiming to reduce their size without compromising performance. Sparsity in neural networks is one…
In the ever-evolving landscape of artificial intelligence (AI), the challenge of creating systems that can effectively collaborate in dynamic environments is a significant one. Multi-agent reinforcement learning (MARL) has been a key focus, aiming to teach agents to interact and adapt in such settings. However, these methods often grapple with complexity and adaptability issues, particularly…
In the domain of sequential decision-making, especially in robotics, agents often deal with continuous action spaces and high-dimensional observations. These difficulties result from making decisions across a broad range of potential actions like complex, continuous action spaces and evaluating enormous volumes of data. Advanced procedures are needed to process and act upon the information in…
Large Language Models (LLMs) face deployment challenges due to latency issues caused by memory bandwidth constraints. Researchers use weight-only quantization to address this, compressing LLM parameters to lower precision. This approach improves latency and reduces GPU memory requirements. Implementing this effectively requires custom mixed-type matrix-multiply kernels that move, dequantize, and process weights efficiently. Existing kernels…
Large Language Models (LLMs) have revolutionized the field of natural language processing, allowing machines to understand and generate human language. These models, such as GPT-4 and Gemini-1.5, are crucial for extensive text processing applications, including summarization and question answering. However, managing long contexts remains challenging due to computational limitations and increased costs. Researchers are, therefore,…
Harvard researchers have recently unveiled ReXrank, an open-source leaderboard dedicated to AI-powered radiology report generation. This significant development is poised to revolutionize the field of healthcare AI, particularly in interpreting chest x-ray images. The introduction of ReXrank aims to set new standards by providing a comprehensive and objective evaluation framework for cutting-edge models. This initiative…
Artificial intelligence, particularly in training large multimodal models (LMMs), relies heavily on vast datasets that include sequences of images and text. These datasets enable the development of sophisticated models capable of understanding and generating multimodal content. As AI models’ capabilities advance, the need for extensive, high-quality datasets becomes even more critical, driving researchers to explore…
Artificial intelligence (AI) is dedicated to developing systems capable of performing tasks that typically require human intelligence. This dedication is met with numerous challenges along the way. One such challenge in AI is creating systems that can manage complex, realistic tasks requiring extensive interaction with dynamic environments. These tasks often involve searching for and synthesizing…