allow setting quantizedLayers of WebGPU mode; chore

2024-03-01 14:23:05 +08:00
parent c9513822c9
commit 887ba06bd6
8 changed files with 46 additions and 10 deletions
--- a/frontend/src/utils/index.tsx
+++ b/frontend/src/utils/index.tsx
@@ -194,6 +194,8 @@ export const getStrategy = (modelConfig: ModelConfig | undefined = undefined) =>
    case 'WebGPU':
    case 'WebGPU (Python)':
      strategy += params.precision === 'nf4' ? 'fp16i4' : params.precision === 'int8' ? 'fp16i8' : 'fp16';
+      if (params.quantizedLayers)
+        strategy += ` layer${params.quantizedLayers}`;
      break;
    case 'CUDA':
    case 'CUDA-Beta':