Skip to content

vllm_omni.transformers_utils.configs.ming_talker

Standalone talker configuration, independent of thinker/vision imports.

MingFlashOmniTalkerConfig

Bases: PretrainedConfig

Configuration class for Ming-flash-omni-2.0 talker (TTS) stage.

The talker uses a Qwen2 LLM backbone with CFM (Conditional Flow Matching) via a DiT diffusion transformer, plus an Aggregator that maps generated audio latents back to the LLM embedding space for autoregressive generation.

aggregator instance-attribute 

aggregator = aggregator or {}

audio_vae_path instance-attribute 

audio_vae_path = audio_vae_path

campplus_model instance-attribute 

campplus_model = campplus_model

cfg_strength instance-attribute 

cfg_strength = cfg_strength

compile_flow instance-attribute 

compile_flow = compile_flow

duration_limit instance-attribute 

duration_limit = duration_limit

enable_llm_cuda_graph instance-attribute 

enable_llm_cuda_graph = enable_llm_cuda_graph

flowmodel instance-attribute 

flowmodel = flowmodel or {}

history_patch_size instance-attribute 

history_patch_size = history_patch_size

latent_dim instance-attribute 

latent_dim = latent_dim

llm_config instance-attribute 

llm_config = llm_config

low_latency_streaming instance-attribute 

low_latency_streaming = low_latency_streaming

max_cache_len instance-attribute 

max_cache_len = max_cache_len

max_decode_steps instance-attribute 

max_decode_steps = max_decode_steps

model_type class-attribute instance-attribute 

model_type = 'ming_flash_omni_talker'

normalize_text instance-attribute 

normalize_text = normalize_text

patch_size instance-attribute 

patch_size = patch_size

scrub_markup instance-attribute 

scrub_markup = scrub_markup

steps instance-attribute 

steps = steps

stream_decode instance-attribute 

stream_decode = stream_decode

strict_weight_loading instance-attribute 

strict_weight_loading = strict_weight_loading

tn_mode instance-attribute 

tn_mode = tn_mode

trim_trailing_silence instance-attribute 

trim_trailing_silence = trim_trailing_silence

get_text_config 

get_text_config(decoder: bool = False) -> PretrainedConfig