Post-training quantization (PTQ) focuses on reducing the size and improving the speed of large language models (LLMs) to make them more practical for real-world use. Such models require large data volumes, but strongly skewed and highly heterogeneous data distribution during quantization presents considerable difficulties. This would inevitably expand the quantization range, making it, in most…
Significant progress has been made in short-form instrumental compositions in AI and music generation. However, creating full songs with lyrics, vocals, and instrumental accompaniment is still challenging for existing models. Generating a full-length song from lyrics poses several challenges. The music is long, requiring AI models to maintain consistency and coherence over several minutes. The…
What is an Agent? An agent is a Large Language Model (LLM)-powered system that can decide its own workflow. Unlike traditional chatbots, which operate on a fixed path (ask → answer), agents are capable of: Choosing between different actions based on context. Using external tools such as web search, databases, or APIs. Looping between steps…
Vision-Language Models (VLMs) have significantly expanded AI’s ability to process multimodal information, yet they face persistent challenges. Proprietary models such as GPT-4V and Gemini-1.5-Pro achieve remarkable performance but lack transparency, limiting their adaptability. Open-source alternatives often struggle to match these models due to constraints in data diversity, training methodologies, and computational resources. Additionally, limited documentation…
Reinforcement learning (RL) trains agents to make sequential decisions by maximizing cumulative rewards. It has diverse applications, including robotics, gaming, and automation, where agents interact with environments to learn optimal behaviors. Traditional RL methods fall into two categories: model-free and model-based approaches. Model-free techniques prioritize simplicity but require extensive training data, while model-based methods introduce…
Large Language Models (LLMs) have emerged as transformative tools in research and industry, with their performance directly correlating to model size. However, training these massive models presents significant challenges, related to computational resources, time, and cost. The training process for state-of-the-art models like Llama 3 405B requires extensive hardware infrastructure, utilizing up to 16,000 H100…
LLMs based on transformer architectures, such as GPT and LLaMA series, have excelled in NLP tasks due to their extensive parameterization and large training datasets. However, research indicates that not all learned parameters are necessary to retain performance, prompting the development of post-training compression techniques to enhance efficiency without significantly reducing inference quality. For example,…