Switch back to subprocessing for llama.cpp

This should resolve a number of memory leak and stability defects by allowing
us to isolate llama.cpp in a separate process and shutdown when idle, and
gracefully restart if it has problems.  This also serves as a first step to be
able to run multiple copies to support multiple models concurrently.

gpu/amd_linux.go

@@ -100,6 +100,8 @@ func AMDGetGPUInfo(resp *GpuInfo) {
 		return
 	}
 
+	updateLibPath(libDir)
+
 	gfxOverride := os.Getenv("HSA_OVERRIDE_GFX_VERSION")
 	if gfxOverride == "" {
 		supported, err := GetSupportedGFX(libDir)
@@ -143,6 +145,21 @@ func AMDGetGPUInfo(resp *GpuInfo) {
 	}
 }
 
+func updateLibPath(libDir string) {
+	ldPaths := []string{}
+	if val, ok := os.LookupEnv("LD_LIBRARY_PATH"); ok {
+		ldPaths = strings.Split(val, ":")
+	}
+	for _, d := range ldPaths {
+		if d == libDir {
+			return
+		}
+	}
+	val := strings.Join(append(ldPaths, libDir), ":")
+	slog.Debug("updated lib path", "LD_LIBRARY_PATH", val)
+	os.Setenv("LD_LIBRARY_PATH", val)
+}
+
 // Walk the sysfs nodes for the available GPUs and gather information from them
 // skipping over any devices in the skip map
 func amdProcMemLookup(resp *GpuInfo, skip map[int]interface{}, ids []int) {

gpu/assets.go

@@ -11,6 +11,7 @@ import (
 	"strings"
 	"sync"
 	"syscall"
+	"time"
 )
 
 var (
@@ -84,7 +85,12 @@ func Cleanup() {
 		slog.Debug("cleaning up", "dir", tmpDir)
 		err := os.RemoveAll(tmpDir)
 		if err != nil {
-			slog.Warn("failed to clean up", "dir", tmpDir, "err", err)
+			// On windows, if we remove too quickly the llama.dll may still be in-use and fail to remove
+			time.Sleep(1000 * time.Millisecond)
+			err = os.RemoveAll(tmpDir)
+			if err != nil {
+				slog.Warn("failed to clean up", "dir", tmpDir, "err", err)
+			}
 		}
 	}
 }