Phi Mini MoE Instruct
Phi Mini MoE Instruct is a compressed mixture-of-experts model with 7.6B total parameters and 2.4B active parameters. It belongs to the SlimMoE line, not the dense Phi-4 Mini line.
What distinguishes this release
Microsoft derives this release from the base shared by Phi-3.5-MoE and GRIN-MoE through SlimMoE compression and distillation. Its card lists a 4K context and English commercial and research use.Phi Mini MoE Instruct official model card
Build the task around its input and output
Practical guidance. For a short-form local task, account for all stored weights when sizing the machine. The active parameter figure describes token computation, not the entire model's storage. Include runtime overhead and simultaneous requests when checking memory use.
Review the result in its intended setting
Practical guidance. Keep retrieval results compact enough for the 4K input budget. A long conversation or a large pasted document can consume the space before the actual question. Compare the Mini and Tiny variants on the short tasks you intend to serve.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Access | Downloadable |
| Model identifiers | microsoft/Phi-mini-MoE-instruct |
| Context window | 4,096 tokens Published checkpoint context; serving engines may impose lower limits Phi Mini MoE Instruct official model card |
| Model size | 7.6B total; 2.4B activePhi Mini MoE Instruct official model card |
| Input and output | Text input and text outputPhi Mini MoE Instruct official model card |
| Weight license | MITPhi Mini MoE Instruct official model card |