vllm_omni.model_executor.models.indextts2.prompt_utils ¶
Prompt helpers for IndexTTS2 talker prefill.
build_indextts2_prefill_prompt_ids ¶
build_indextts2_prefill_prompt_ids(
model_id_or_path: str,
text: str,
*,
placeholder_token_id: int = 1,
) -> list[int]
estimate_indextts2_prefill_prompt_len ¶
Return the placeholder prompt length expected by the IndexTTS2 talker.
The model-side prefill replaces placeholder embeddings with: 34 conditioning tokens, text tokens wrapped with start/stop markers, and one start-mel token.