Skip to content

vllm_omni.transformers_utils.configs.vibevoice

VibeVoice config for the native HF-schema checkpoint (vibevoice/VibeVoice-1.5B-hf).

The project's Transformers 5.13/5.14 dependency lacks the full TTS config, so this shim provides the native schema without loading remote code. Only the native sub-config layout is accepted: audio_config, semantic_model_config, text_config and diffusion_head_config. The original Microsoft schema and the early fork layout are not converted.

VibeVoiceConfig

Bases: PreTrainedConfig

Root config for the native VibeVoice checkpoint.

audio_bos_token_id instance-attribute 

audio_bos_token_id = audio_bos_token_id

audio_config instance-attribute 

audio_config = _sub_config(
    audio_config, "vibevoice_acoustic_tokenizer"
)

audio_eos_token_id instance-attribute 

audio_eos_token_id = audio_eos_token_id

audio_token_id instance-attribute 

audio_token_id = audio_token_id

diffusion_head_config instance-attribute 

diffusion_head_config = diffusion_head_config

hidden_size property 

hidden_size: int

model_type class-attribute instance-attribute 

model_type = 'vibevoice'

semantic_model_config instance-attribute 

semantic_model_config = _sub_config(
    semantic_model_config,
    "vibevoice_acoustic_tokenizer_encoder",
    hidden_size=128,
)

text_config instance-attribute 

text_config = _sub_config(text_config, 'qwen2')

tie_word_embeddings instance-attribute 

tie_word_embeddings = getattr(
    self.text_config, "tie_word_embeddings", False
)

vocab_size instance-attribute 

vocab_size = self.text_config.vocab_size

VibeVoiceDiffusionHeadConfig

Bases: PreTrainedConfig

Native-schema diffusion head sub-config (no model_type lookup).

diffusion_max_period instance-attribute 

diffusion_max_period = diffusion_max_period

frequency_embedding_size instance-attribute 

frequency_embedding_size = frequency_embedding_size

hidden_act instance-attribute 

hidden_act = hidden_act

hidden_size instance-attribute 

hidden_size = hidden_size

intermediate_size instance-attribute 

intermediate_size = intermediate_size

latent_size instance-attribute 

latent_size = latent_size

mlp_bias instance-attribute 

mlp_bias = mlp_bias

model_type class-attribute instance-attribute 

model_type = 'vibevoice_diffusion'

num_hidden_layers instance-attribute 

num_hidden_layers = num_hidden_layers

rms_norm_eps instance-attribute 

rms_norm_eps = rms_norm_eps