DeepYard
M

Maple

Efficient mixture-of-experts reasoning model with ternary quantization for multi-hop tasks

Open SourceFree

About

Maple is an open-source reasoning model from DeepGrove that uses a novel mixture-of-experts architecture with ternary quantization to enable efficient multi-hop reasoning for AI agents. The preview release demonstrates how MoE techniques can reduce computational overhead while maintaining reasoning quality across complex agent tasks. Optimized for scenarios requiring multiple inference steps and logical chains.

Details

Typecoding-agent
Deploymentself-hosted
Supported Models

Tags

open-sourceautonomouspythonreasoningmulti-agent