From c63bb1d16a70c03440671b76954bb767513cead8 Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?Johannes=20G=C3=A4=C3=9Fler?= Date: Tue, 22 Aug 2023 22:47:05 +0200 Subject: CUDA: use mul_mat_q kernels by default (#2683) --- common/common.h | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'common/common.h') diff --git a/common/common.h b/common/common.h index c50a6edf..18fd951e 100644 --- a/common/common.h +++ b/common/common.h @@ -68,7 +68,7 @@ struct gpt_params { size_t hellaswag_tasks = 400; // number of tasks to use when computing the HellaSwag score bool low_vram = false; // if true, reduce VRAM usage at the cost of performance - bool mul_mat_q = false; // if true, use experimental mul_mat_q kernels + bool mul_mat_q = true; // if true, use mul_mat_q kernels instead of cuBLAS bool memory_f16 = true; // use f16 instead of f32 for memory kv bool random_prompt = false; // do not randomize prompt if none provided bool use_color = false; // use color to distinguish generations and inputs -- cgit v1.2.3