Phi-4-mini-reasoning — GGUF (iPhone-optimized)
A GGUF quantization of microsoft/Phi-4-mini-reasoning for on-device inference on iPhone, iPad, and Apple Silicon Macs via llama.cpp or apps that wrap it (e.g. Haplo).
Hosted by jc-builds for the Haplo ecosystem. Original weights © Microsoft Corporation, redistributed under the MIT License per the upstream license.
TL;DR
The reasoning-tuned sibling of Phi-4-mini-instruct: same 3.8B dense architecture, further trained on synthetic chain-of-thought math and logic data. It works problems out step by step before answering, which makes replies longer but far more reliable on multi-step math. MIT-licensed. See the upstream model card for benchmarks.
Available quantizations
| File | Size | Recommended use |
|---|---|---|
Phi-4-mini-reasoning-Q4_K_M.gguf |
2.49 GB | Default — best size/quality tradeoff for phone & laptop |
Details
| Parameters | 3.8B (dense) |
| Context window | 128K tokens |
| Architecture | phi3 (same as Phi-4-mini-instruct) |
| Quantization | Q4_K_M |
| Minimum device | 8 GB RAM (iPhone 15 Pro / 16 / 17 class), iPad, Mac |
How to use
Haplo (iPhone / iPad / Mac)
The model appears automatically in Haplo's model browser. Download URL:
https://huggingface.co/jc-builds/Phi-4-mini-reasoning-GGUF/resolve/main/Phi-4-mini-reasoning-Q4_K_M.gguf
llama.cpp
llama-cli -hf jc-builds/Phi-4-mini-reasoning-GGUF:Q4_K_M
License
MIT. Original model © Microsoft Corporation — see the upstream license.