ernie4_5_moe_gguf¶


Model type	`ernie4_5_moe_gguf`
Class	`Ernie45MoEGGUFCausalLMModel`
Task	`text-generation`
Source	`models/moe.py`

Description¶

ERNIE 4.5 MoE graph matching llama.cpp’s scheduled GGUF semantics.

Dense prefix and periodic routed layers are selected from GGUF metadata. Routed layers use bias-corrected softmax selection, unbiased normalized weights, and an optional always-active shared SwiGLU expert.

Usage¶

mobius build --model <MODEL_ID> output_dir/

from mobius import build

model = build("<MODEL_ID>")