transformer_lens.model_bridge.supported_architectures.qwen2_audio module

Qwen2-Audio architecture adapter.

Qwen2AudioForConditionalGeneration: a Whisper-style audio encoder at model.audio_tower, a linear projector at model.multi_modal_projector, and a Qwen2 text decoder at model.language_model with a top-level lm_head (the transformers >= 5.13 layout). Text-only forwards work without audio; audio features enter via input_features.

class transformer_lens.model_bridge.supported_architectures.qwen2_audio.Qwen2AudioArchitectureAdapter(cfg: Any)

Bases: Qwen2ArchitectureAdapter

Architecture adapter for Qwen2AudioForConditionalGeneration models.

__init__(cfg: Any) → None

Initialize the Qwen2-Audio architecture adapter.