diff --git a/LLama/Extensions/IContextParamsExtensions.cs b/LLama/Extensions/IContextParamsExtensions.cs index 4f922890d..469d0517c 100644 --- a/LLama/Extensions/IContextParamsExtensions.cs +++ b/LLama/Extensions/IContextParamsExtensions.cs @@ -58,8 +58,7 @@ public static void ToLlamaContextParams(this IContextParams @params, out LLamaCo null => LLamaFlashAttentionType.LLAMA_FLASH_ATTENTION_TYPE_AUTO }; result.kv_unified = true; - result.n_seq_max = (uint)Math.Min(Math.Max(10, result.n_ctx / 8), 64); - + result.n_threads = Threads(@params.Threads); result.n_threads_batch = Threads(@params.BatchThreads);