dbrx_gguf¶


Model type	`dbrx_gguf`
Class	`DbrxGGUFCausalLMModel`
Task	`text-generation`
Source	`models/moe.py`

Description¶

DBRX causal LM graph matching llama.cpp’s GGUF inference semantics.

Each block uses weight-only LayerNorm, one fused QKV projection clamped before its [Q | K | V] split, and full-softmax top-k routed SwiGLU experts whose selected weights are renormalized.

Usage¶

mobius build --model <MODEL_ID> output_dir/

from mobius import build

model = build("<MODEL_ID>")