diff --git a/llm/llama.go b/llm/llama.go
index 0b460e9..b79e04a 100644
--- a/llm/llama.go
+++ b/llm/llama.go
@@ -299,10 +299,6 @@ func newLlama(model string, adapters []string, runners []ModelRunner, numLayers
 		params = append(params, "--n-gpu-layers", fmt.Sprintf("%d", numGPU))
 	}
 
-	if opts.NumGQA > 0 {
-		params = append(params, "--gqa", fmt.Sprintf("%d", opts.NumGQA))
-	}
-
 	if len(adapters) > 0 {
 		// TODO: applying multiple adapters is not supported by the llama.cpp server yet
 		params = append(params, "--lora", adapters[0])