refactor: consolidate eval routes into cforch.py #44

New issue

Closed

opened 2026-05-01 12:12:01 -07:00 by pyr0ball · 1 comment

pyr0ball commented

2026-05-01 12:12:01 -07:00

Owner

Context

Four separate FastAPI route modules (style.py, plans_bench.py, voice.py, plus classifier routes in api.py) each have their own model-loading and reporting logic. All benchmarks should route through cf-orch so they exercise real end-user paths.

Work

Absorb routes from app/style.py, app/plans_bench.py, app/voice.py into app/eval/cforch.py
Remove direct model calls; all benchmarks go through cf-orch allocate → run → release
Standalone CLI scripts (benchmark_style.py, benchmark_voice.py, etc.) survive as manage.sh commands — only their API routes are absorbed
Remove now-empty style.py, plans_bench.py, voice.py app modules

Acceptance

./manage.sh style-bench, voice-bench, plans-bench still work
All benchmark results available via /api/eval/* routes
No direct model connections from avocet API layer

## Context Four separate FastAPI route modules (`style.py`, `plans_bench.py`, `voice.py`, plus classifier routes in `api.py`) each have their own model-loading and reporting logic. All benchmarks should route through cf-orch so they exercise real end-user paths. ## Work - Absorb routes from `app/style.py`, `app/plans_bench.py`, `app/voice.py` into `app/eval/cforch.py` - Remove direct model calls; all benchmarks go through cf-orch allocate → run → release - Standalone CLI scripts (`benchmark_style.py`, `benchmark_voice.py`, etc.) survive as `manage.sh` commands — only their API routes are absorbed - Remove now-empty `style.py`, `plans_bench.py`, `voice.py` app modules ## Acceptance - `./manage.sh style-bench`, `voice-bench`, `plans-bench` still work - All benchmark results available via `/api/eval/*` routes - No direct model connections from avocet API layer