vh/esh-pfi-infrastructure
1
0
Fork 0
Code Issues 7 Pull Requests Actions Packages Projects Releases Wiki Activity

stacks/kokoro: flip .env.example default from cpu → gpu (driver bump landed)

Browse Source 
irv-ml1's driver upgrade to 595.58.03 (kernel 6.1.0-37, CUDA 13.2) is
working — both GPUs detected, modules loaded. The gpu variant of the
Kokoro-FastAPI image (which requires CUDA >= 12.9) is now the right
default for new deploys. Flipping KOKORO_VARIANT=gpu, KOKORO_USE_GPU=true,
KOKORO_GPU_DEVICES=0 (pins to the RTX 3090 — Kokoro is ~1 GB VRAM and
doesn't need the A6000).
This commit is contained in:
Vuong Hoang
2026-04-27 15:17:27 -07:00
parent 5ef502e7cc
commit e0a33d0b33
1 changed files with 10 additions and 12 deletions
Download Patch File Download Diff File Expand all files Collapse all files
stacks/kokoro/.env.example
+10 -12
View File
@@ -2,14 +2,13 @@
# ── image pin ────────────────────────────────────────────────────────
# Image variant. Two choices:
# cpu — works on any host. Kokoro is tiny (82M params) so CPU
# inference is acceptable: TTFA ~1s on a modern CPU vs
# ~300ms on a 30-series-or-newer GPU. Fine as a default.
# gpu — requires NVIDIA driver supporting CUDA >= 12.9. As of
# 2026-04-25, irv-ml1's driver 570.124.06 caps at CUDA 12.8,
# so the gpu variant won't start there yet. Bump the driver
# (separate procedure, requires reboot) then flip this to gpu.
KOKORO_VARIANT=cpu
# gpu — default. Requires NVIDIA driver supporting CUDA >= 12.9.
# irv-ml1's driver bumped 2026-04-25 from 570.124.06 → 595.58.03
# (CUDA 13.2), so the gpu variant runs there. ~300ms TTFA.
# cpu — fallback for any host without GPU passthrough. Kokoro is
# tiny (82M params) so CPU inference is workable: TTFA ~1s vs
# ~300ms on GPU.
KOKORO_VARIANT=gpu
# Tagged release on GHCR. Avoid `latest` — upstream warns it can move
# without notice. v0.2.4-master = 2025-12-13 release with Kokoro-82M v1.0
@@ -30,12 +29,11 @@ KOKORO_BIND=0.0.0.0
# Only used when KOKORO_VARIANT=gpu. Leave empty for the cpu variant.
# When using the gpu variant: "0" pins to the RTX 3090 (Kokoro is tiny
# — ~1 GB VRAM — and doesn't need the A6000); "all" exposes both.
KOKORO_GPU_DEVICES=
KOKORO_GPU_DEVICES=0
# Tells the wrapper to route inference through CUDA. Must be `false`
# for the cpu variant (it has no CUDA libraries). Flip to `true` when
# you switch KOKORO_VARIANT=gpu.
KOKORO_USE_GPU=false
# for the cpu variant (it has no CUDA libraries) and `true` for gpu.
KOKORO_USE_GPU=true
# Logging level for the FastAPI app. INFO is the upstream default.
KOKORO_LOG_LEVEL=INFO