Most teams scale inference by adding more GPUs. Smart teams think about how tokens flow. 😎 Introducing fastokens — Crusoe's open-source Rust tokenizer, built with NVIDIA Dynamo, now merged into SGLang. Up to 50% faster TTFT for agentic workloads, measured on real production https://t.co/RkEYx9UO3h
May 6, 2026
Views13.2k
Comments3
Reposts3