Skip to content

feat: add managed llama.cpp audio runtime - #7

Draft
Nothing1596 wants to merge 33 commits into
feat/sensory-audio-inferencefrom
feat/sensory-managed-llama
Draft

Nothing1596 wants to merge 33 commits into
feat/sensory-audio-inferencefrom
feat/sensory-managed-llama

Conversation

@Nothing1596

Copy link
Copy Markdown
Owner

Summary

  • Stack PR 6/6, split from upstream feat: add sensory framework Rvosy/Sakura#99.
  • Adds the optional managed llama.cpp audio runtime path, runtime/model manifests, deployment checks, doctor CLI, one-click prepare flow, and all-source smoke planning.
  • Keeps large downloads explicit and guards smoke tests that would trigger remote GGUF downloads.

Base

Validation

  • Full test suite on this stack branch: 1403 passed.
  • Final stack content matches original feat/sensory-framework-clean head exactly.

@Nothing1596

Copy link
Copy Markdown
Owner Author

本轮按“llama.cpp 只是增强功能,不做独立产品”的约束收敛:

  • 设置页主入口改成“准备本机音频增强”,诊断摘要第一行直接显示“已准备/未准备”。
  • managed llama 首次启动加锁,避免并发 observe 重复启动 sidecar。
  • tar 解压从 fully_trusted 改为更保守的 filter="data",保留原有路径/link 校验。
  • 新增高级维护命令 cleanup-cache,默认 dry-run,--yes 才清理 Sakura 管理的 runtime 子目录和推荐 speech/sound 音频模型缓存;不删除 manifest、手动 root binary 或视觉模型缓存。

验证:

  • py_compile: audio_runtime_cli / llama_cpp_runtime / providers / settings_dialog / sections 通过
  • tests/unit/test_sensory_audio_runtime_cli.py tests/unit/test_sensory_llama_cpp_runtime.py tests/unit/test_sensory.py: 110 passed
  • tests/ui/test_pet_window.py: 263 passed
  • focused combined: 373 passed
  • full pytest -q: 1408 passed

@Nothing1596

Copy link
Copy Markdown
Owner Author

根据复审补了两个非阻塞风险收口:

  • cleanup-cache --target runtime --yes 现在跳过 archives/_downloads/ 和点开头 staging 目录,只清理已安装 runtime 子目录,避免离线 manifest 指向的 archive 被误删。
  • 文档补充发布前门禁:macOS arm64 / Windows x64 / Linux x64 各跑一次真实官方包安装解压,确认 filter="data" 兼容;随后 speech/sound 各跑一次真实 smoke --allow-model-download。同时提示运行中 sidecar 在 Windows 上可能导致清理失败,需要先关闭或重启后再清理。

验证:

  • tests/unit/test_sensory_audio_runtime_cli.py: 36 passed
  • focused managed llama/UI set: 373 passed
  • full pytest -q: 1408 passed

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant