Skip to content

vllm_omni.transformers_utils.configs.audio8_tts

Audio8 TTS config registration with transformers AutoConfig.

Registers model_type = "arktts" (and the two sub-config types) so AutoConfig.from_pretrained("Audio8/Audio8-TTS-Preview-0.6b") resolves to vllm-omni's Qwen2-shaped config instead of the checkpoint's remote code.

This only wins when trust_remote_code is off -- transformers prefers a checkpoint's auto_map over registered classes otherwise -- which is why deploy/audio8_tts.yaml sets trust_remote_code: false.

Audio8TTSConfig

Bases: PretrainedConfig

Top-level Audio8 TTS config (model_type = "arktts").

Accepts the flat HF checkpoint fields and derives text_config (Slow AR) and fast_ar_config (Fast AR). get_text_config() returns the Slow AR config, which is what Qwen2Model reads.

codebook_size instance-attribute 

codebook_size = int(codebook_size)

codec_filename instance-attribute 

codec_filename = str(codec_filename)

codec_frame_size instance-attribute 

codec_frame_size = int(codec_frame_size)

codec_post_intermediate_size instance-attribute 

codec_post_intermediate_size = int(
    codec_post_intermediate_size
)

codec_post_n_head instance-attribute 

codec_post_n_head = int(codec_post_n_head)

codec_post_n_layer instance-attribute 

codec_post_n_layer = int(codec_post_n_layer)

codec_post_n_local_heads instance-attribute 

codec_post_n_local_heads = int(codec_post_n_local_heads)

codec_sample_rate instance-attribute 

codec_sample_rate = int(codec_sample_rate)

fast_ar_config instance-attribute 

fast_ar_config = fast_ar_config or Audio8TTSFastARConfig(
    codebook_size=codebook_size,
    num_codebooks=num_codebooks,
    fast_dim=fast_dim,
    fast_n_head=fast_n_head,
    fast_n_local_heads=fast_n_local_heads,
    fast_head_dim=fast_head_dim,
    n_fast_layer=n_fast_layer,
    fast_intermediate_size=fast_intermediate_size,
    fast_attention_qkv_bias=fast_attention_qkv_bias,
    fast_attention_qk_norm=fast_attention_qk_norm,
    rope_base=rope_base,
    norm_eps=norm_eps,
)

model_type class-attribute instance-attribute 

model_type = 'arktts'

norm_fastlayer_input instance-attribute 

norm_fastlayer_input = bool(norm_fastlayer_input)

num_codebooks instance-attribute 

num_codebooks = int(num_codebooks)

ras_temperature instance-attribute 

ras_temperature = float(ras_temperature)

ras_top_p instance-attribute 

ras_top_p = float(ras_top_p)

ras_window_size instance-attribute 

ras_window_size = int(ras_window_size)

semantic_begin_id instance-attribute 

semantic_begin_id = int(semantic_begin_id)

semantic_end_id instance-attribute 

semantic_end_id = int(semantic_end_id)

sub_configs class-attribute instance-attribute 

sub_configs = {
    "text_config": Audio8TTSSlowARConfig,
    "fast_ar_config": Audio8TTSFastARConfig,
}

text_config instance-attribute 

text_config = text_config or Audio8TTSSlowARConfig(
    vocab_size=vocab_size,
    dim=dim,
    n_head=n_head,
    n_local_heads=n_local_heads,
    head_dim=head_dim,
    n_layer=n_layer,
    intermediate_size=intermediate_size,
    max_seq_len=max_seq_len,
    rope_base=rope_base,
    norm_eps=norm_eps,
    attention_qkv_bias=attention_qkv_bias,
    attention_qk_norm=attention_qk_norm,
    tie_word_embeddings=tie_word_embeddings,
    codebook_size=codebook_size,
    num_codebooks=num_codebooks,
    semantic_begin_id=semantic_begin_id,
    semantic_end_id=semantic_end_id,
    eos_token_id=eos_token_id,
    pad_token_id=pad_token_id,
)

get_text_config 

get_text_config(
    *args: Any, **kwargs: Any
) -> Audio8TTSSlowARConfig

Audio8TTSFastARConfig

Bases: PretrainedConfig

Fast AR config: the n_fast_layer residual-codebook predictor.

attention_qk_norm instance-attribute 

attention_qk_norm = bool(fast_attention_qk_norm)

attention_qkv_bias instance-attribute 

attention_qkv_bias = bool(fast_attention_qkv_bias)

head_dim instance-attribute 

head_dim = int(fast_head_dim)

hidden_act instance-attribute 

hidden_act = 'silu'

hidden_size instance-attribute 

hidden_size = int(fast_dim)

intermediate_size instance-attribute 

intermediate_size = int(fast_intermediate_size)

max_position_embeddings instance-attribute 

max_position_embeddings = int(num_codebooks)

model_type class-attribute instance-attribute 

model_type = 'arktts_fast_ar'

num_attention_heads instance-attribute 

num_attention_heads = int(fast_n_head)

num_codebooks instance-attribute 

num_codebooks = int(num_codebooks)

num_hidden_layers instance-attribute 

num_hidden_layers = int(n_fast_layer)

num_key_value_heads instance-attribute 

num_key_value_heads = int(fast_n_local_heads)

rms_norm_eps instance-attribute 

rms_norm_eps = float(norm_eps)

rope_theta instance-attribute 

rope_theta = float(rope_base)

Audio8TTSSlowARConfig

Bases: PretrainedConfig

Slow AR config exposed with Qwen2-compatible attribute names.

attention_qk_norm instance-attribute 

attention_qk_norm = bool(attention_qk_norm)

attention_qkv_bias instance-attribute 

attention_qkv_bias = bool(attention_qkv_bias)

codebook_size instance-attribute 

codebook_size = int(codebook_size)

head_dim instance-attribute 

head_dim = int(head_dim)

hidden_act instance-attribute 

hidden_act = 'silu'

hidden_size instance-attribute 

hidden_size = int(dim)

intermediate_size instance-attribute 

intermediate_size = int(intermediate_size)

max_position_embeddings instance-attribute 

max_position_embeddings = int(max_seq_len)

model_type class-attribute instance-attribute 

model_type = 'arktts_slow_ar'

num_attention_heads instance-attribute 

num_attention_heads = int(n_head)

num_codebooks instance-attribute 

num_codebooks = int(num_codebooks)

num_hidden_layers instance-attribute 

num_hidden_layers = int(n_layer)

num_key_value_heads instance-attribute 

num_key_value_heads = int(n_local_heads)

rms_norm_eps instance-attribute 

rms_norm_eps = float(norm_eps)

rope_parameters instance-attribute 

rope_parameters = {
    "rope_type": "default",
    "rope_theta": float(rope_base),
}

semantic_begin_id instance-attribute 

semantic_begin_id = int(semantic_begin_id)

semantic_end_id instance-attribute 

semantic_end_id = int(semantic_end_id)