Research papers

AI Hardware & Efficiency

Supercharging AI: How ‘LLM in a Flash’ Revolutionizes Language Model Inference on Memory-Limited Devices
ByAjith Vallath Prabhakar January 15, 2024November 20, 2025

Large Language Models (LLMs) have impressive natural language processing capabilities, but they require significant computational resources. Apple’s “LLM in a flash” solution overcomes this challenge by using flash memory to store model parameters, reducing data transfers, and optimizing memory efficiency. This breakthrough allows advanced language models to operate on devices with limited memory, making AI more accessible.

Read More Supercharging AI: How ‘LLM in a Flash’ Revolutionizes Language Model Inference on Memory-Limited Devices
Reasoning Systems & Multimodal AI

OneLLM: One Framework to Align All Modalities with Language
ByAjith Vallath Prabhakar January 1, 2024November 20, 2025

Multimodal Large Language Models (MLLMs) have the ability to process information from different sensory modalities. However, current MLLMs are facing several challenges such as complex integration, scalability issues, high resource requirements, and increased risk of overfitting. To overcome these challenges, researchers have developed OneLLM, which is a revolutionary MLLM that aligns eight different modalities of language using a unified framework. OneLLM has a simplified architecture and reduced resource requirements, allowing it to use a single structure for various modalities. This feature facilitates increased task versatility, enhanced cross-modal comprehension, and broader application scope across different industries.

Read More OneLLM: One Framework to Align All Modalities with Language
AI Models & Architectures

LLM360: Fully Transparent Open-Source LLMs
ByAjith Vallath Prabhakar December 18, 2023February 16, 2025

Transparency plays a crucial role in the development of Large Language Models (LLMs). It promotes ethical AI development, encourages innovation, and maintains scientific integrity. One noteworthy initiative in this regard is LLM360, which aims to achieve complete transparency in LLM training. This initiative addresses significant challenges related to data provenance, reproducibility, and open collaboration. LLM360 promotes transparency by open-sourcing its training data, codes, and checkpoints, which allows for widespread study, replication, and innovation of advanced LLMs.

Read More LLM360: Fully Transparent Open-Source LLMs