34 lines
979 B
Go
34 lines
979 B
Go
package llamacpp
|
|
|
|
import (
|
|
"context"
|
|
"fmt"
|
|
"net/http"
|
|
"path/filepath"
|
|
|
|
"github.com/docker/model-runner/pkg/logging"
|
|
)
|
|
|
|
func (l *llamaCpp) ensureLatestLlamaCpp(ctx context.Context, log logging.Logger, httpClient *http.Client,
|
|
llamaCppPath, vendoredServerStoragePath string,
|
|
) error {
|
|
nvGPUInfoBin := filepath.Join(vendoredServerStoragePath, "com.docker.nv-gpu-info.exe")
|
|
var canUseCUDA11 bool
|
|
var err error
|
|
if ShouldUseGPUVariant {
|
|
canUseCUDA11, err = hasCUDA11CapableGPU(ctx, nvGPUInfoBin)
|
|
if err != nil {
|
|
l.status = fmt.Sprintf("failed to check CUDA 11 capability: %v", err)
|
|
return fmt.Errorf("failed to check CUDA 11 capability: %w", err)
|
|
}
|
|
}
|
|
desiredVersion := "latest"
|
|
desiredVariant := "cpu"
|
|
if canUseCUDA11 {
|
|
desiredVariant = "cuda"
|
|
}
|
|
l.status = fmt.Sprintf("looking for updates for %s variant", desiredVariant)
|
|
return l.downloadLatestLlamaCpp(ctx, log, httpClient, llamaCppPath, vendoredServerStoragePath, desiredVersion,
|
|
desiredVariant)
|
|
}
|