inference: Fix nv-gpu-info path and wrap errors

Signed-off-by: Piotr Stankiewicz <[email protected]>

Author: Piotr Stankiewicz

pkg/gpuinfo/memory_windows.go

@@ -20,7 +20,7 @@ func getVRAMSize(modelRuntimeInstallPath string) (uint64, error) {
 		return 1, nil
 	}
 
-	nvGPUInfoBin := filepath.Join(modelRuntimeInstallPath, "com.docker.nv-gpu-info.exe")
+	nvGPUInfoBin := filepath.Join(modelRuntimeInstallPath, "bin", "com.docker.nv-gpu-info.exe")
 
 	ctx, _ := context.WithTimeout(context.Background(), 30*time.Second)
 	cmd := exec.CommandContext(ctx, nvGPUInfoBin)

pkg/inference/backends/llamacpp/llamacpp.go

@@ -226,19 +226,19 @@ func (l *llamaCpp) GetDiskUsage() (int64, error) {
 func (l *llamaCpp) GetRequiredMemoryForModel(model string, config *inference.BackendConfiguration) (*inference.RequiredMemory, error) {
 	mdl, err := l.modelManager.GetModel(model)
 	if err != nil {
-		return nil, err
+		return nil, fmt.Errorf("getting model(%s): %w", model, err)
 	}
 	mdlPath, err := mdl.GGUFPath()
 	if err != nil {
-		return nil, err
+		return nil, fmt.Errorf("getting gguf path for model(%s): %w", model, err)
 	}
 	mdlGguf, err := parser.ParseGGUFFile(mdlPath)
 	if err != nil {
-		return nil, err
+		return nil, fmt.Errorf("parsing gguf(%s): %w", mdlPath, err)
 	}
 	mdlConfig, err := mdl.Config()
 	if err != nil {
-		return nil, err
+		return nil, fmt.Errorf("accessing model(%s) config: %w", model, err)
 	}
 
 	contextSize := GetContextSize(&mdlConfig, config)