llm: normalise kvct parameter handling (#7926)

llm/memory.go

@@ -129,7 +129,7 @@ func EstimateGPULayers(gpus []discover.GpuInfo, ggml *GGML, projectors []string,
 
 	var kvct string
 	if fa {
-		requested := envconfig.KvCacheType()
+		requested := strings.ToLower(envconfig.KvCacheType())
 		if requested != "" && ggml.SupportsKVCacheType(requested) {
 			kvct = requested
 		}

llm/server.go

@@ -225,7 +225,7 @@ func NewLlamaServer(gpus discover.GpuInfoList, model string, ggml *GGML, adapter
 		fa = false
 	}
 
-	kvct := envconfig.KvCacheType()
+	kvct := strings.ToLower(envconfig.KvCacheType())
 
 	if fa {
 		slog.Info("enabling flash attention")