Sakana AI and Nvidia Unveil TwELL Sparse LLM Solution
SakanaAILabs · x · 2026-07-04
Sakana AI and Nvidia have introduced the TwELL sparse packing format. It adapts to optimized chunked GPU tasks and features custom CUDA kernels to accelerate LLM training and inference. Training and benchmarking on billion-parameter sparse models show a speedup of over 20%, alongside even greater reductions in peak memory and power consumption. The paper will be presented at ICML 2026.
More from Infra
- SmolVM open-sources persistent computer infrastructure for agents that outlive chat sessions — aniketmaurya · 2026-09-11
- PyTorch Day Korea 2026 launches first offline conf, CFP closes Sept 13 — PyTorch · 2026-09-11
- Local LLM server dilemma: 4x CMP-170HX (price up 53% in 20 days) vs Mac Studio M5 Ultra — rumboll · 2026-09-11
- llama.cpp lands Flash Attention tuning for RDNA4, big prefill gains on AMD — pmttyji · 2026-09-11
- Your p99 latency benchmark may be lying: a deep dive into coordinated omission — Franc0Fernand0 · 2026-09-11
- Running MiniMax H3 on 12GB VRAM: quantization, Turbo LoRAs and attention backends compared — Possible_Mood676 · 2026-09-11