58 lines
1.6 KiB
YAML
58 lines
1.6 KiB
YAML
# MOSS-TTS-Realtime (MossTTSRealtime, 1.7B, n_vq=16, 24 kHz)
|
|
# HF: OpenMOSS-Team/MOSS-TTS-Realtime
|
|
# Real-time streaming TTS with TTFB ~180 ms and LLM+TTS first-sentence ~377 ms.
|
|
# Uses local depth transformer (no delay warm-up) — lowest latency variant.
|
|
async_chunk: true
|
|
# OpenMOSS-Team/MOSS-TTS-Realtime ships custom modeling code; the newer vLLM
|
|
# ModelConfig hard-rejects it at validation without this (was lenient before).
|
|
trust_remote_code: true
|
|
|
|
connectors:
|
|
shm:
|
|
name: SharedMemoryConnector
|
|
extra:
|
|
codec_streaming: true
|
|
connector_get_sleep_s: 0.005
|
|
connector_get_max_wait_first_chunk: 1000
|
|
connector_get_max_wait: 300
|
|
codec_chunk_frames: 15 # smaller chunks for lower TTFA
|
|
codec_left_context_frames: 0
|
|
|
|
stages:
|
|
- stage_id: 0
|
|
max_num_seqs: 8
|
|
gpu_memory_utilization: 0.60
|
|
enable_prefix_caching: false
|
|
async_scheduling: true
|
|
max_num_batched_tokens: 4096
|
|
max_model_len: 4096
|
|
devices: "0"
|
|
output_connectors:
|
|
to_stage_1: shm
|
|
default_sampling_params:
|
|
temperature: 1.0
|
|
top_p: 0.95
|
|
top_k: 50
|
|
max_tokens: 4096
|
|
seed: 42
|
|
repetition_penalty: 1.1
|
|
|
|
- stage_id: 1
|
|
max_num_seqs: 1
|
|
gpu_memory_utilization: 0.12
|
|
enforce_eager: true
|
|
enable_prefix_caching: false
|
|
async_scheduling: true
|
|
max_num_batched_tokens: 65536
|
|
max_model_len: 65536
|
|
devices: "0"
|
|
input_connectors:
|
|
from_stage_0: shm
|
|
default_sampling_params:
|
|
temperature: 0.0
|
|
top_p: 1.0
|
|
top_k: -1
|
|
max_tokens: 65536
|
|
seed: 42
|
|
repetition_penalty: 1.0
|