Support multiple variants for a given llm lib type

In some cases we may want multiple variants for a given GPU type or CPU.
This adds logic to have an optional Variant which we can use to select
an optimal library, but also allows us to try multiple variants in case
some fail to load.

This can be useful for scenarios such as ROCm v5 vs v6 incompatibility
or potentially CPU features.
This commit is contained in:
Daniel Hiltgen
2024-01-05 12:13:08 -08:00
parent b24e8d17b2
commit 8da7bef05f
16 changed files with 428 additions and 212 deletions

View File

@@ -2,9 +2,6 @@ package llm
import (
"embed"
"errors"
"fmt"
"io/fs"
"log"
"os"
"strings"
@@ -24,23 +21,3 @@ func updatePath(dir string) {
log.Printf("Updating PATH to %s", newPath)
os.Setenv("PATH", newPath)
}
func verifyDriverAccess() error {
// Only check ROCm access if we have the dynamic lib loaded
if _, rocmPresent := AvailableShims["rocm"]; rocmPresent {
// Verify we have permissions - either running as root, or we have group access to the driver
fd, err := os.OpenFile("/dev/kfd", os.O_RDWR, 0666)
if err != nil {
if errors.Is(err, fs.ErrPermission) {
return fmt.Errorf("Radeon card detected, but permissions not set up properly. Either run ollama as root, or add you user account to the render group.")
} else if errors.Is(err, fs.ErrNotExist) {
// expected behavior without a radeon card
return nil
}
return fmt.Errorf("failed to check permission on /dev/kfd: %w", err)
}
fd.Close()
}
return nil
}