Skip to content

knowledge: daily intake batch 05 #127

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
tzhouam merged 1 commit into from
Sep 5, 2026
Merged

knowledge: daily intake batch 05 #127

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
7 changes: 7 additions & 0 deletions knowledge/repos/vllm-omni/models/dots-tts/rules.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -34,3 +34,10 @@ confidence: high
hardware marker；weekly `-m "slow and L4 and tts"` 收集 3，nightly
`-m "full_model and L4 and tts"` 收集 0。PR 未在 L4 实跑，真实 weekly run 才能证明硬件结果。
^[PR #6174]

## DOTS-2a — online serving detection 必须以 dots.tts architecture 优先，不能认领共享 stage key

- 触发：修改 `dots_tts` 的 `/v1/audio/speech` adapter、TTS detector metadata、`serving_speech` 的 stage discovery，或与 VoxCPM2 共用 `latent_generator` 一类 stage key 的模型接入。
- 强制：`DotsTTSAdapter` 只按 `DotsTTSForConditionalGeneration` architecture 识别 dots.tts，并用高于 VoxCPM2 的 `detect_priority` 先参与排序；`stage_keys` 必须保持空集合，避免把共享的 `latent_generator` entry stage 当作 dots.tts 的充分证据。最终 online speech 路由必须经统一 adapter detection，而不是在 `serving_speech.py` 增加模型专用分支。
- 禁止：让 dots.tts 通过共享 stage key 抢占 VoxCPM2 请求；在优先级上落后于 VoxCPM2 再依赖 stage 扫描“碰运气”命中；或把一次成功的 dots.tts speech 请求当成 detector 消歧已被证明。
- 验收：检测测试必须同时覆盖 dots.tts architecture 命中 `DotsTTSAdapter`、VoxCPM2 architecture 仍命中 `VoxCPM2Adapter`，并断言 dots.tts 不因 `latent_generator` 进入候选；serving 集成测试还要证明 `/v1/audio/speech` 的 dots.tts 请求走到正确 adapter。E2E 通过只证明该精确 online-serving 路径可达，不外推为其他共享 stage-key 模型都已安全。^[PR #6235]