longcat_flash¶


Model type	`longcat_flash`
Class	`LongcatFlashCausalLMModel`
Task	`text-generation`
Source	`models/longcat_flash.py`

Description¶

LongCat Flash causal LM with dual-sublayer MLA + MoE architecture.

Each physical decoder layer contains two MLA attention sub-layers, two dense MLPs, and one MoE shortcut block computed once per layer.

model_type: longcat_flash

mobius build --model <MODEL_ID> output_dir/

from mobius import build

model = build("<MODEL_ID>")