For those interested in the implementation details, we’ve open-sourced the reference code for this paper. The repository includes our sparse training code and the custom CUDA kernels designed for H100 GPUs leveraging the TwELL packing format. GitHub:
AI HARDWARE
-
Running Speech Recognition and LLaMA-GPT on Raspberry Pi
By
–
Running Speech Recognition and a LLaMA-GPT on a Raspberry Pi! #BigData #Analytics #DataScience #AI #MachineLearning #NLProc #IoT #IIoT #PyTorch #Python #RStats #TensorFlow #Java #JavaScript #ReactJS #GoLang #CloudComputing #Serverless #DataScientist #Linux #Programming #Coding
-
IndiaAI Mission Initiatives Aim to Reduce AI Compute Costs
By
–
@SecretaryMEITY, highlights how the IndiaAI Mission is making AI compute more affordable—bringing costs down to nearly one-third of current market rates and expanding access to AI innovation across the Global South.#IndiaAI #DigitalIndia #AIforAll #AICompute #GlobalSouth… pic.twitter.com/TGv0lT3Wm1
— IndiaAI (@OfficialINDIAai) 9 mai 2026@SecretaryMEITY
, highlights how the IndiaAI Mission is making AI compute more affordable—bringing costs down to nearly one-third of current market rates and expanding access to AI innovation across the Global South. #IndiaAI #DigitalIndia #AIforAll #AICompute #GlobalSouth -
Strategic necessity of domestic AI hardware manufacturing
By
–
Président @EmmanuelMacron si j'étais vous je donnerais immédiatement l'ordre de créer une grande usine de composants électroniques de dernière génération pour les besoins de l’IA sur le territoire français. C'est déjà trop tard mais il n'est jamais trop tard pour bien faire.
-
Sakana AI Accelerates Sparse LLMs with NVIDIA
By
–
Sakana AIは、@NVIDIAとの共同研究で、スパースなTransformer言語モデルの推論・学習を高速化する新しいGPUカーネルとデータ形式を開発しました。
— Sakana AI (@SakanaAILabs) 9 mai 2026
ブログ：https://t.co/fMARMRFsJJ
LLMのコストの大部分を占めるフィードフォワード層では、実は各トークンに対して大半の活性がほぼゼロで無駄な計算に… https://t.co/nTMg0QgdSrSakana AI has developed new GPU kernels and data formats that accelerate inference and training of sparse Transformer language models through joint research with @NVIDIA
. Blog: https://
pub.sakana.ai/sparser-faster
-llms/
… In the feedforward layers, which account for the majority of LLM costs, most -
The essential need for GPU clusters in AI
By
–
It is a truth universally acknowledged, that a single man in possession of a good fortune, must be in want of a GPU cluster.
-
Running 30B models on laptop reduces battery life significantly
By
–
A lot. I've run ~30B models on a laptop on a plane and noticed that the battery life was considerably reduced, though I didn't record exact numbers
-
GPUs anciens pourraient faire tourner l’AGI
By
–
Could you imagine that AGI might run on GPUs from 10+ years ago? Because it just might There’s a reason I haven’t sold a single one of my GPUs despite the massive jump in price Compute is the MOAT + you don’t know what the next paradigm might allow you to do with thet hardware
-
Hardware limitations on sparse neural activation in LLMs
By
–
The human brain is incredibly efficient because it only activates the specific neurons needed for a thought. Modern LLMs naturally try to do this too (> 95% of neurons in feedforward layers stay silent for any given word), but our hardware punishes them for it. One of the most
-
Optimiser les LLM avec la sparsité adaptée au GPU
By
–
How do we make LLMs faster and lighter? Don’t force the GPU to adapt to sparsity. Reshape the sparsity to fit the GPU! ⚡️
— Sakana AI (@SakanaAILabs) 8 mai 2026
Excited to share our new #ICML2026 paper in collaboration with @NVIDIA: "Sparser, Faster, Lighter Transformer Language Models". This work introduces new… pic.twitter.com/ehByWHIh6IHow do we make LLMs faster and lighter? Don’t force the GPU to adapt to sparsity. Reshape the sparsity to fit the GPU! Excited to share our new #ICML2026 paper in collaboration with @NVIDIA
: "Sparser, Faster, Lighter Transformer Language Models". This work introduces new