Research & Papers · Other Companies

Top 10 KV Cache Compression Techniques for LLM Inference: Reducing Memory Overhead Across Eviction, Quantization, and Low-Rank Methods
Research & Papers

Top 10 KV Cache Compression Techniques for LLM Inference: Reducing Memory Overhead Across Eviction, Quantization, and Low-Rank Methods

The article discusses 10 KV cache compression techniques for optimizing Large Language Model inference, covering...

MarkTechPost
The MIT-IBM Computing Research Lab launches to shape the future of AI and quantum computing
Research & Papers

The MIT-IBM Computing Research Lab launches to shape the future of AI and quantum computing

MIT and IBM have launched a new computing research lab focused on the convergence of AI, algorithms, and quantum...

MIT News AI
smol-audio: A Colab-Friendly Notebook Collection for Fine-Tuning Whisper, Parakeet, Voxtral, Granite Speech, and Audio Flamingo 3
Research & Papers

smol-audio: A Colab-Friendly Notebook Collection for Fine-Tuning Whisper, Parakeet, Voxtral, Granite Speech, and Audio Flamingo 3

smol-audio is a collection of Google Colab-friendly notebooks designed for fine-tuning multiple audio AI models...

MarkTechPost
Enabling privacy-preserving AI training on everyday devices
Research & Papers

Enabling privacy-preserving AI training on everyday devices

A new method enables privacy-preserving AI training on everyday devices, making accurate and efficient AI models...

MIT News AI
The evolution of encoders: From simple models to multimodal AI
Research & Papers

The evolution of encoders: From simple models to multimodal AI

The article explores encoders as a fundamental component of AI systems that convert real-world information into...

AI News
Researchers find AI text is making the internet more uniform and weirdly cheerful
Research & Papers

Researchers find AI text is making the internet more uniform and weirdly cheerful

A large-scale analysis of Internet Archive websites reveals that AI-generated text is already widespread on the web,...

The Decoder
Top 10 Physical AI Models Powering Real-World Robots in 2026
Research & Papers

Top 10 Physical AI Models Powering Real-World Robots in 2026

A new class of foundation models purpose-built for physical action rather than text generation is now being deployed on...

MarkTechPost
How to Build a Lightweight Vision-Language-Action-Inspired Embodied Agent with Latent World Modeling and Model Predictive Control
Research & Papers

How to Build a Lightweight Vision-Language-Action-Inspired Embodied Agent with Latent World Modeling and Model Predictive Control

This tutorial demonstrates how to build a lightweight embodied AI agent that learns to perceive, plan, and act directly...

MarkTechPost
Meet Talkie-1930: A 13B Open-Weight LLM Trained on Pre-1931 English Text for Historical Reasoning and Generalization Research
Research & Papers

Meet Talkie-1930: A 13B Open-Weight LLM Trained on Pre-1931 English Text for Historical Reasoning and Generalization Research

Researchers have developed Talkie-1930, a 13 billion parameter open-weight language model trained exclusively on...

MarkTechPost
A faster way to estimate AI power consumption
Research & Papers

A faster way to estimate AI power consumption

EnergAIzer is a new method that quickly estimates AI power consumption in seconds, providing reliable results for data...

MIT News AI
How to Build Smarter Multilingual Text Wrapping with BudouX Through Parsing, HTML Rendering, Model Introspection, and Toy Training
Research & Papers

How to Build Smarter Multilingual Text Wrapping with BudouX Through Parsing, HTML Rendering, Model Introspection, and Toy Training

This tutorial explores BudouX, a library for intelligent phrase-aware line breaking in languages without natural...

MarkTechPost
RAG Without Vectors: How PageIndex Retrieves by Reasoning
Research & Papers

RAG Without Vectors: How PageIndex Retrieves by Reasoning

The article discusses PageIndex, a RAG (Retrieval-Augmented Generation) system that retrieves information through...

MarkTechPost
MIT scientists build the world’s largest collection of Olympiad-level math problems, and open it to everyone
Research & Papers

MIT scientists build the world’s largest collection of Olympiad-level math problems, and open it to everyone

MIT scientists have created the world's largest collection of over 30,000 Olympiad-level math problems from 47...

MIT News AI
AI galaxy hunters are adding to the global GPU crunch
Research & Papers

AI galaxy hunters are adding to the global GPU crunch

Astronomers are utilizing GPUs to identify distant galaxies in astronomical data, contributing to the growing global...

TechCrunch AI
A Detailed Implementation on Equinox with JAX Native Modules, Filtered Transforms, Stateful Layers, and End-to-End Training Workflows
Research & Papers

A Detailed Implementation on Equinox with JAX Native Modules, Filtered Transforms, Stateful Layers, and End-to-End Training Workflows

This tutorial explores Equinox, a JAX-based neural network library, demonstrating how to use its PyTree-based module...

MarkTechPost
Watch Sony’s elite ping-pong robot beat top-ranked players
Research & Papers

Watch Sony’s elite ping-pong robot beat top-ranked players

Sony AI has developed Ace, a ping-pong robot that can compete against and occasionally defeat top-ranked human players...

The Verge AI
Decoupled DiLoCo: A new frontier for resilient, distributed AI training
Research & Papers

Decoupled DiLoCo: A new frontier for resilient, distributed AI training

Decoupled DiLoCo presents a new approach to distributed AI training that improves resilience and fault tolerance in...

DeepMind Blog
A Coding Implementation to Build a Conditional Bayesian Hyperparameter Optimization Pipeline with Hyperopt, TPE, and Early Stopping
Research & Papers

A Coding Implementation to Build a Conditional Bayesian Hyperparameter Optimization Pipeline with Hyperopt, TPE, and Early Stopping

This tutorial demonstrates how to implement an advanced Bayesian hyperparameter optimization workflow using Hyperopt...

MarkTechPost
Moonshot AI and Tsinghua Researchers Propose PrfaaS: A Cross-Datacenter KVCache Architecture that Rethinks How LLMs are Served at Scale
Research & Papers

Moonshot AI and Tsinghua Researchers Propose PrfaaS: A Cross-Datacenter KVCache Architecture that Rethinks How LLMs are Served at Scale

Moonshot AI and Tsinghua University researchers propose PrfaaS, a novel cross-datacenter KVCache architecture that...

MarkTechPost