From e0a33d0b33e7475cb0296c45dc93500ce6045a2b Mon Sep 17 00:00:00 2001 From: Vuong Hoang Date: Mon, 27 Apr 2026 15:17:27 -0700 Subject: [PATCH] =?UTF-8?q?stacks/kokoro:=20flip=20.env.example=20default?= =?UTF-8?q?=20from=20cpu=20=E2=86=92=20gpu=20(driver=20bump=20landed)?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit 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). --- stacks/kokoro/.env.example | 22 ++++++++++------------ 1 file changed, 10 insertions(+), 12 deletions(-) diff --git a/stacks/kokoro/.env.example b/stacks/kokoro/.env.example index 8b5a190..9258447 100644 --- a/stacks/kokoro/.env.example +++ b/stacks/kokoro/.env.example @@ -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