feat(openwhispr): wire real CUDA support for local transcription
CI / lint (push) Successful in 27s
CI / eval (push) Successful in 1m50s

The onnxruntime CUDA execution provider it bundles was hard-failing back
to CPU: konishi has the driver but no CUDA toolkit runtime libs. Adds an
opt-in cudaSupport toggle (cudaPackages_13, pinned to match the provider's
build) and enables it on konishi, the only host with an NVIDIA GPU.
This commit is contained in:
Miguel Palhas
2026-09-07 16:28:00 +01:00
parent dfe8d1648f
commit a3f06bcd24
4 changed files with 59 additions and 9 deletions
+1
View File
@@ -15,6 +15,7 @@ let
};
openwhispr = pkgs.callPackage ../../../pkgs/openwhispr/package.nix {
inherit (config.custom.aiApps) deviceScaleFactor;
cudaSupport = config.custom.aiApps.cudaAcceleration;
};
in
{
+8
View File
@@ -19,6 +19,14 @@
description = "--force-device-scale-factor value for Claude Desktop and T3 Code.";
};
# OpenWhispr's GPU-accelerated local Whisper transcription (NVIDIA hosts
# only). Pulls in cudaPackages_13 (multi-GB) when enabled, so opt-in per host.
options.custom.aiApps.cudaAcceleration = lib.mkOption {
type = lib.types.bool;
default = false;
description = "Wire OpenWhispr's onnxruntime CUDA execution provider against cudaPackages_13.";
};
# ~/.claude/settings.json stays unmanaged: Claude Code rewrites it itself
# (model pins, permission grants, plugin state), so any nix copy drifts within
# a session and every `nh home switch` then aborts on the diff.