flan
b18b0ddd90
adscrub transcribe: downloads audio, transcribes episodes with no chapter-sourced ad spans, stores segment timestamps. Device picked at runtime via ctranslate2.get_cuda_device_count() -- CUDA if visible, CPU int8 otherwise -- so the same code works from a plain dev shell or a GPU-enabled Docker deploy. Added compose.gpu.yaml to request the host's GPU via the nvidia Docker runtime and build with the cublas/cudnn extra. Corrected CLAUDE.md/docs/PLAN.md: code has a real RTX 2070 SUPER and Docker's nvidia runtime is registered -- the earlier 'no GPU' note only reflected this interactive shell's LXC lacking device passthrough, not the host's actual capability.
19 lines
598 B
YAML
19 lines
598 B
YAML
# GPU override: requests the host's RTX 2070 SUPER via the nvidia Docker runtime
|
|
# (confirmed registered on `code`, 2026-07-10 — see CLAUDE.md) and builds the image
|
|
# with the cuBLAS/cuDNN extra so faster-whisper actually uses it.
|
|
#
|
|
# docker compose -f compose.yaml -f compose.gpu.yaml build
|
|
# docker compose -f compose.yaml -f compose.gpu.yaml run --rm adscrub transcribe
|
|
services:
|
|
adscrub:
|
|
build:
|
|
args:
|
|
GPU: "1"
|
|
deploy:
|
|
resources:
|
|
reservations:
|
|
devices:
|
|
- driver: nvidia
|
|
count: 1
|
|
capabilities: ["gpu"]
|