DeepYardDeepYard
I

Instella MoE

AMD's 16B parameter MoE model optimized for agentic reasoning on ROCm hardware

Open SourceFree

About

Instella MoE is AMD's open-source mixture-of-experts language model featuring 16 billion parameters with 3 billion active experts per token. Specifically optimized for AMD's ROCm platform, it delivers enhanced reasoning capabilities for agentic workflows while maintaining computational efficiency through its sparse expert activation. Designed as a foundation for building autonomous agents that require strong logical reasoning and decision-making abilities.

Details

Typecoding-agent
Deploymentself-hosted
Supported Models

Tags

open-sourceautonomousself-hostedcoding-agentpython