InData Science CollectivebySagar Ganapaneni·Nov 1Quantizing an Analytics Agent Without Losing the DecisionA response icon1A response icon1
Hannah Igboke·Oct 26How to Make a 14GB Model Lose WeightCan Mr Jumbo really lose weight and have have a normal life? Let’s find out.A response icon1A response icon1
Rafał Łagowski·Oct 26Reducir un modelo de mantenimiento predictivo para que quepa en nueve dispositivos de borde…Hace poco encontré un artículo de Oliveira et al. que compara métodos de compresión de modelos en muchos tipos de hardware. Los autores los…
Thara·Oct 25The Smaller AI Model Hiding Inside a Bigger OneA farmer notices strange spots on a crop’s leaves. They take a photo, open an app, and get a prediction about the disease.A response icon1A response icon1
Abioye Oluwadamilola Joy·Oct 9Why Shrinking AI for Edge Devices Can Go Wrong (And What I Learned)Everyone is talking about putting artificial intelligence directly onto tiny devices things like smart thermostats, farm sensors, or…
Manjeetkumar·Sep 29Model Compression Series: The Mathematics Behind LoRA, Quantization and PruningHow large AI models are made smaller and faster, from SVD and low-rank updates to GPTQ, QLoRA and KV cache compression.
Anandha L Ranganathan·Sep 24Model Compression with QuantizationQuantization is one of the most important model compression techniques used to make neural networks smaller, faster, and more…
Alejandro Mata Ali·Sep 18Comprimir LLMs con Tensor Networks: por qué reducir parámetros no bastaLos grandes modelos de lenguaje son grandes en un sentido bastante literal. Buena parte de lo que han aprendido está almacenado en enormes…
Cyrus·Aug 26Qwen3.6 Model Surgery: 35B-A3B to 27B-A2.8BHow I shrank a 35B MoE to 27B, healed it with a $53 LoRA run, and got a 15–30% decode speedup on laptop hardware.
InGoPenAIbyRimshaqaism·Jul 29The One Concept That Explains Why AI Models Can Be CompressedI’ve been studying the architecture of large models — how GPT and Claude actually work under the hood. That led me down a rabbit hole: how…
