phimoe_gguf¶


Model type	`phimoe_gguf`
Class	`PhiMoEGGUFCausalLMModel`
Task	`text-generation`
Source	`models/moe.py`

Description¶

PhiMoE graph matching llama.cpp’s GGUF inference semantics.

The HuggingFace model uses SparseMixer routing, while the pinned llama.cpp loader uses full softmax followed by normalized top-k routing. GGUF import therefore uses this internal graph without changing native HF builds.

Usage¶

mobius build --model <MODEL_ID> output_dir/

from mobius import build

model = build("<MODEL_ID>")