qwen3_5_text¶


Model type	`qwen3_5_text`
Class	`Qwen35CausalLMModel`
Task	`hybrid-text-generation`
Source	`models/qwen35.py`

Description¶

Qwen3.5 causal language model with hybrid linear/full attention.

Combines standard GQA layers (with output gating) and GatedDeltaNet linear attention layers in a single decoder stack. The per-layer attention type is controlled by config.layer_types.

Full attention layers use standard KV cache. DeltaNet layers carry conv_state and recurrent_state tensors, managed by

class:: HybridCausalLMTask.

Usage¶

mobius build --model <MODEL_ID> output_dir/

from mobius import build

model = build("<MODEL_ID>")