local-v15-streaming
Operate MOSS-TTS Local Transformer v1.5 batch and realtime streaming decode workflows.
- License 许可
- Apache-2.0
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- local-v15-streaming
- license
- Apache 2.0
- metadata
- { "disco-role": "operating" }
- description
- Operate MOSS-TTS Local Transformer v1.5 batch and realtime streaming decode workflows.
- disable-model-invocation
- true
02 资源文件 4
- 参考文档references/api-reference.md9.6 KB
- 参考文档references/batch-and-streaming.md8.6 KB
- 参考文档references/troubleshooting.md7.8 KB
- 脚本scripts/estimate_local_v15_tokens.py7.5 KB
03 SKILL.md 正文
local-v15-streaming
Use this sub-skill when the task is about MOSS-TTS-Local-Transformer-v1.5 local batch inference, realtime streaming decode, browser streaming app operation, 48 kHz stereo codec-v2 audio, device/dtype split, token or duration estimates, continuation prompts, or voice-clone prompts.
Route elsewhere
- MOSS-TTS-Realtime voice-agent model, realtime conversation, or live voice-agent serving:
../realtime-voice-agent/SKILL.md. - Generic Hugging Face family prompts, non-v1.5 model-family usage, or router-level model choice:
../hf-family-workflows/SKILL.md. - Fine-tuning, dataset preparation, manifests, or training local v1.5:
../finetuning-data-prep/SKILL.md. - SoundEffect v2 generation or diffusion sound-effect workflows:
../soundeffect-v2/SKILL.md.
Minimum operating facts
- Checkpoint:
OpenMOSS-Team/MOSS-TTS-Local-Transformer-v1.5. - Codec/tokenizer:
OpenMOSS-Team/MOSS-Audio-Tokenizer-v2. - Audio format: 48 kHz stereo; decoded tensors are normally
[2, samples]. - Frame topology: 12.5 frames/sec, 12 RVQ layers, time-synchronous local-transformer frames.
- Backbone: Qwen3-4B-derived local transformer release.
- Default audio sampling for the app: temperature
1.7, top-p0.8, top-k25, repetition penalty1.0.
Use the bundled references
- Batch inference, web streaming launch, mode semantics, and prompt patterns:
references/batch-and-streaming.md. - Python/CLI/app API fields and event contracts:
references/api-reference.md. - Failure modes and fixes:
references/troubleshooting.md. - Safe no-model estimate helper:
scripts/estimate_local_v15_tokens.py.
Before claiming a v1.5 workflow is correct, verify that the runtime reports the expected sample rate (48000), RVQ depth (12), and stereo channel count, and that final output includes a WAV, audio-token dump, and metadata JSON in the configured output location.
04 同仓库更多 Skills
passnet-triton-opt
Make a single PassNet Triton kernel fast AND numerically faithful: performance model (when a replacement can win at all), block/grid/warp tuning, autotune policy, launch overhead, and per-op numeric recipes to pass the dtype baseline tolerances. Use when a pass matches and is correct but speedup below expected, or when correctness fails by small numeric margins.
passnet-solve
END-TO-END playbook for solving one PassNet sample: analyze the computation graphs, decide the optimization strategy (what to fuse, what to replace, what to leave alone), drive the iteration loop, and maximize the sample score. This is the ENTRY skill — invoke it first for any PassNet optimization task; it tells you when to use passnet-pattern-fusion, passnet-triton-opt and passnet-feedback.
passnet-skill
PassNet GPU kernel optimization via compiler passes. Design and implement Triton-based optimization passes, create pass files under ./pass_dir/, self-evaluate with pass_evaluator, and iterate to maximize GPU speedup.
passnet-pattern-fusion
Author PassNet pass files: write patterns that actually MATCH the FX graph, pick fusion regions (which ops to absorb into one Triton kernel), and structure multi-pass submissions with the mandatory shared-dispatch architecture. Use when creating or fixing pattern()/replacement_args()/replacement_func() files, when a pass "failed to match", or when deciding how to fuse multiple kernels/ops.
passnet-orchestrate
Round/batch planning and decision support for PassNet graph optimization. Use for multi-sample triage, worker allocation, eval-budget policy, and post-eval keep/revert decisions. For solving one concrete sample end to end, use passnet-solve as the entry skill; consult this skill only when you need broader planning or a specific decision gate.
passnet-feedback
Fast, deterministic feedback for PassNet work: pre-flight pattern/match verification WITHOUT burning a GPU evaluation, per-node bottleneck analysis of a sample's graphs, and eval-log parsing into per-variant status + estimated score + failure classification. Use BEFORE every GPU evaluation (check_pattern), at the START of a sample (analyze_graph), and AFTER every evaluation (parse_eval_log).