SAAYN Bundle Manifest

Smoke prompt: Mike is quick. He moves

Expected smoke token output: quickly

Expected optimized engine: spreadsheet-gpt2-numpy

| Bundle Variant | Archive | Size | Includes Workbook | Includes CSV Params | Includes .venv | Offline Runnable | SHA256 |
|---|---:|---:|---:|---:|---:|---:|---|
| Ultra-light source | saayn-openai-bundle-ultralight-source.zip | 2.3M | No | No | No | No | e516283869dcef326c23855b26c63528c03988b27c7587c96b807b2b44419927 |
| Lite | saayn-openai-bundle-lite.zip | 577M | No | Yes | No | No | a28f9892882cf2f22b1e695c01bf5a1c8b8720192a26fc1f02482de1e704afc9 |
| Full | saayn-openai-bundle-full.zip | 1.9G | Yes | Yes | No | No | 52159a2a161a819b559f1dbc226258f2d0f127c28662b1aa2e65f2d35ed4384f |
| Runtime Linux | saayn-openai-bundle-runtime-linux.zip | 2.5G | Yes | Yes | Yes | Yes, on similar Linux/Python 3.10 x86_64 | f62f93dc85dbf6dde744b10f0c64822162f56de42ac2262568f54dbf57ecb7ea |

Notes:

• Ultra-light downloads Python dependencies and GPT-2 CSV parameters during bash bundle_scripts/install.sh.
• Lite and full include model artifacts, but still need dependency installation unless the target environment already has them.
• Runtime Linux is the only archive intended to run without downloading or compiling first on a compatible machine.
• The workbook is optional for runtime; the optimized backend uses the CSV parameters and NumPy cache headlessly.
• Chat mode defaults to dialogue prompting. Use --normal-completion, /completion on, or /mode completion for plain current-input completion with no dialogue primer, no session-history prompt, no role labels, no injected Assistant:, and no dialogue stop markers.