mirror of https://github.com/ollama/ollama.git
57 lines
1.7 KiB
Go
57 lines
1.7 KiB
Go
//go:build linux || windows
|
|
|
|
package discover
|
|
|
|
import (
|
|
"fmt"
|
|
"log/slog"
|
|
"os"
|
|
"regexp"
|
|
"runtime"
|
|
"strconv"
|
|
"strings"
|
|
)
|
|
|
|
// Jetson devices have JETSON_JETPACK="x.y.z" factory set to the Jetpack version installed.
|
|
// Included to drive logic for reducing Ollama-allocated overhead on L4T/Jetson devices.
|
|
var CudaTegra string = os.Getenv("JETSON_JETPACK")
|
|
|
|
func cudaVariant(gpuInfo CudaGPUInfo) string {
|
|
if runtime.GOARCH == "arm64" && runtime.GOOS == "linux" {
|
|
if CudaTegra != "" {
|
|
ver := strings.Split(CudaTegra, ".")
|
|
if len(ver) > 0 {
|
|
return "jetpack" + ver[0]
|
|
}
|
|
} else if data, err := os.ReadFile("/etc/nv_tegra_release"); err == nil {
|
|
r := regexp.MustCompile(` R(\d+) `)
|
|
m := r.FindSubmatch(data)
|
|
if len(m) != 2 {
|
|
slog.Info("Unexpected format for /etc/nv_tegra_release. Set JETSON_JETPACK to select version")
|
|
} else {
|
|
if l4t, err := strconv.Atoi(string(m[1])); err == nil {
|
|
// Note: mapping from L4t -> JP is inconsistent (can't just subtract 30)
|
|
// https://developer.nvidia.com/embedded/jetpack-archive
|
|
switch l4t {
|
|
case 35:
|
|
return "jetpack5"
|
|
case 36:
|
|
return "jetpack6"
|
|
default:
|
|
slog.Info("unsupported L4T version", "nv_tegra_release", string(data))
|
|
}
|
|
}
|
|
}
|
|
}
|
|
return "sbsa"
|
|
}
|
|
|
|
// driver 12.0 has problems with the cuda v12 library, so run v11 on those older drivers
|
|
if gpuInfo.DriverMajor < 12 || (gpuInfo.DriverMajor == 12 && gpuInfo.DriverMinor == 0) {
|
|
// The detected driver is older than Feb 2023
|
|
slog.Warn("old CUDA driver detected - please upgrade to a newer driver", "version", fmt.Sprintf("%d.%d", gpuInfo.DriverMajor, gpuInfo.DriverMinor))
|
|
return "v11"
|
|
}
|
|
return "v12"
|
|
}
|