Zebra-HyLo

mentions 1 type Organization feed RSS

// recent coverage 1 mentions

00:00

2026-09-22

rocm.blogs.amd.com

large-language-models

Zebra-HyLo: Upcycling Transformers into Long-Context Hybrid LLMs on AMD Instinct™ GPUs

AMD researchers introduced Zebra-HyLo, an upcycling recipe that converts pretrained Transformer models into long-context hybrid LLMs through post-training on roughly 10B tokens, without pretraining fr…

// co-occurs with top 7 entities

AMD 1 Zebra-Llama 1 Llama-3.2-1B 1 Llama-3.2-3B 1 Qwen3-1.7B 1 AMD Instinct MI300X 1 vLLM 1

// topics top 5 topics

large language models 1 ai research 1 ai infrastructure 1 ai chips 1 machine learning 1