Zyphra releases ZAYA1-8B: an MoE model trained entirely on AMD clusters
Original title:ZAYA1-8B: Frontier intelligence density, trained on AMD
AI summary
Zyphra releases ZAYA1-8B, an MoE model with fewer than 1 billion active parameters and the first model to complete pretraining, mid-training and supervised fine-tuning entirely on an AMD Instinct MI300 cluster (1,024 nodes, built with IBM).
Key Takeaways
- Zyphra releases ZAYA1-8B
- an MoE model with fewer than 1 billion active parameters and the first model to complete pretraining
- mid-training and supervised fine-tuning entirely on an AMD Instinct MI300 cluster (1,024…
Quick Facts
- Published
- Updated
- Source
- Zyphra Research(网页)
- Category
- Model Releases
- AI score
- 67
原文给出活跃参数不足 1B 的 MoE 模型在数学和编码基准上的具体成绩,以及全 AMD 集群训练和 Markovian RSA 测试时计算方法。
Source & copyright
This article is based on curated titles and AI summaries from the AIHOT public API, with attribution and links back. Original copyright remains with the original publishers and authors; this site does not mirror full texts.
How to Cite
When citing this article, keep the title, date, and permalink, and credit Orange News as the source.
Orange News: Zyphra releases ZAYA1-8B: an MoE model trained entirely on AMD clusters. May 6, 2026. https://new.nat6.net/en/news/cmtym34cd000fro1n2yr52in4/