2026-05-08 14:12:23 +08:00
|
|
|
# OmniVoice TTS deploy: diffusion engine single stage (dit).
|
|
|
|
|
# Text → iterative unmasking → codebook tokens → DAC decode → 24kHz audio.
|
|
|
|
|
# (Migrated from model_executor/stage_configs/omnivoice.yaml for deploy/ layout.)
|
2026-04-03 16:41:20 -04:00
|
|
|
|
|
|
|
|
stage_args:
|
|
|
|
|
- stage_id: 0
|
|
|
|
|
stage_type: diffusion
|
|
|
|
|
is_comprehension: true
|
|
|
|
|
runtime:
|
|
|
|
|
devices: 0
|
|
|
|
|
engine_args:
|
|
|
|
|
model_stage: dit
|
|
|
|
|
model_class_name: "OmniVoicePipeline"
|
|
|
|
|
enforce_eager: true
|
|
|
|
|
trust_remote_code: true
|
|
|
|
|
distributed_executor_backend: "mp"
|
|
|
|
|
dtype: "float32"
|
|
|
|
|
final_output: true
|
|
|
|
|
final_output_type: audio
|
