diff --git a/src/services/llm_local.rs b/src/services/llm_local.rs index 746d367..349b3a4 100644 --- a/src/services/llm_local.rs +++ b/src/services/llm_local.rs @@ -187,7 +187,7 @@ async fn start_llm_server( let mut cmd = tokio::process::Command::new("sh"); cmd.arg("-c").arg(format!( - "cd {} && ./llama-server -m {} --host 0.0.0.0 --port {} --threads 8 --threads-batch 18 --temp 0.7 --parallel 3 --repeat-penalty 1.1 --ctx-size 2096 --batch-size 2048 -n 255 --mlock --no-mmap &", + "cd {} && ./llama-server -m {} --host 0.0.0.0 --port {} --threads 8 --threads-batch 18 --temp 0.7 --parallel 3 --repeat-penalty 1.1 --ctx-size 1024 --batch-size 4096 -n 255 --mlock --no-mmap &", llama_cpp_path, model_path, port ));