JSLMPR 8c0ddc1796 perf(vision): add resident llama-server backend (model loads once) ... The CLI backend reloads the ~3GB model per frame (~2min/frame). Add a SERVER mode: when LLAMACPP_SERVER_BIN is set, start llama-server once, POST base64 frames to its OpenAI /v1/chat/completions endpoint on loopback, stop it at the end. Same answers, ~3x faster on a full clip (measured: 3 frames 181s incl. one-time load vs ~6-8min). CLI mode (LLAMACPP_MTMD_BIN) remains the simple fallback. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>		2026-07-26 17:49:36 +02:00
..
beat_detect.py	feat(cinematic): R10–R14 grade/cadence/edit rules + 420p compat fix	2026-07-25 12:36:18 +02:00
local_asset_requirements.txt	add asset generation	2026-07-21 10:03:27 +02:00
local_asset_worker.py	Wire local generative audio models into the highlight asset worker	2026-07-21 23:00:57 +02:00
local_cv_requirements.txt	Add runnable local CV model worker	2026-07-11 16:40:59 +02:00
local_cv_worker.py	Add rich video editing diagnostics	2026-07-11 17:11:35 +02:00
provision_local_models.py	Wire local generative audio models into the highlight asset worker	2026-07-21 23:00:57 +02:00
run_local_asset_worker.sh	add asset generation	2026-07-21 10:03:27 +02:00
run_local_cv_worker.sh	add asset generation	2026-07-21 10:03:27 +02:00
subject_track.py	feat(cinematic): R10–R14 grade/cadence/edit rules + 420p compat fix	2026-07-25 12:36:18 +02:00
vision_caption.py	Automatic two-tier highlight director + source-adaptive rendering	2026-07-23 22:49:05 +02:00
vision_caption_llamacpp.py	perf(vision): add resident llama-server backend (model loads once)	2026-07-26 17:49:36 +02:00