Support Local Small Language Models for Reasoning and Decision Optimization

[ctava-msft]

Summary

Allow Agent Learning to leverage local SLMs for reasoning, scoring, decision support, and action selection.

Candidate Models
Microsoft Phi family
Future MAI models
Open-weight local reasoning models
Proposed Work
Local inference abstraction
Offline execution mode
Quantization guidance
Pluggable model architecture
Acceptance Criteria
Local model interface implemented
Phi integration sample available
Offline mode validated