Expand description
Application profiles — how inference is used, not only which GPU path.
Timothy (2026-07-10): on a local device, work need not be live. High-stakes multi-system health eval / differential analysis can run overnight and deliver a verified HTML (or email body). Different applications want different modes:
| Profile | Latency | Mid-decode | Post-turn | Decode budget | Timeout |
|---|---|---|---|---|---|
| Interactive | low | optional | light | 256 | 30s |
| LiveFast | lowest | off (FastVerify) | graph heal | 256 | 30s |
| BatchOvernight | irrelevant | off | full HTML+CML | 2048 | 8h |
No Ollama API: all profiles stay in-process Qualia (P64 + resident GEMV + graph).
Enums§
Functions§
- active_
application_ profile - apply_
application_ profile - bootstrap_
application_ profile - Bootstrap from env
QUALIA_APP_PROFILE(call early with path selector). - set_
application_ profile - Apply profile: inference mode, budgets, timeouts, HTML return for batch.