Skip to content

vllm_omni.model_executor.models.audex.audex_code2wav

Audex code2wav: streaming speech-codec-token → waveform stage.

Wraps the vendored Audex causal speech decoder (see speech_decoder/). Unlike the flow/vocoder stages of other TTS models, the decoder is natively stateful and streaming: one AudexCausalSpeechDecoderSession per request holds the decode cache, each engine step pushes only the newly received codec frames, and the session's 4-frame lookahead tail is drained by flush() when the request finishes. The stage therefore never re-decodes left context and always returns delta audio.

logger module-attribute 

logger = init_logger(__name__)

AudexCode2Wav

Bases: Module

Stage-1 model for Audex TTS (GenerationModelRunner).

decoder instance-attribute 

decoder = AudexCausalSpeechDecoderModel(config)

enable_update_additional_information instance-attribute 

enable_update_additional_information = True

has_postprocess instance-attribute 

has_postprocess = False

has_preprocess instance-attribute 

has_preprocess = False

have_multimodal_outputs instance-attribute 

have_multimodal_outputs = True

input_modalities class-attribute instance-attribute 

input_modalities = 'audio'

model_path instance-attribute 

model_path = vllm_config.model_config.model

requires_raw_input_tokens instance-attribute 

requires_raw_input_tokens = True

vllm_config instance-attribute 

vllm_config = vllm_config

compute_logits 

compute_logits(
    hidden_states: Tensor | OmniOutput,
    sampling_metadata: Any = None,
) -> None

embed_input_ids 

embed_input_ids(input_ids: Tensor, **_: Any) -> Tensor

forward 

forward(
    input_ids: Tensor | None = None,
    positions: Tensor | None = None,
    intermediate_tensors: Any = None,
    inputs_embeds: Tensor | None = None,
    runtime_additional_information: list[dict[str, Any]]
    | None = None,
    **kwargs: Any,
) -> OmniOutput

load_weights 

load_weights(
    weights: Iterable[tuple[str, Tensor]],
) -> set[str]

make_omni_output 

make_omni_output(
    model_outputs: Tensor | OmniOutput | tuple,
    **kwargs: Any,
) -> OmniOutput

on_requests_finished 

on_requests_finished(
    finished_req_ids: Iterable[str],
) -> None

Free decoder sessions for finished/aborted requests.