From e77094faab2723f7b9a1ca591a78162b98c56ef8 Mon Sep 17 00:00:00 2001 From: Andrew Godfrey Date: Wed, 15 Nov 2023 18:17:11 -0800 Subject: [PATCH 1/3] Fix #4017 --- ggml-cuda.cu | 4 ++++ 1 file changed, 4 insertions(+) diff --git a/ggml-cuda.cu b/ggml-cuda.cu index f87f18802c8f8..58e7ecb2ed40d 100644 --- a/ggml-cuda.cu +++ b/ggml-cuda.cu @@ -6251,6 +6251,8 @@ static int64_t get_row_rounding(ggml_type type) { return max_compute_capability >= CC_RDNA2 ? 128 : 64; case GGML_TYPE_F16: return 1; + case GGML_TYPE_F32: + return 1; case GGML_TYPE_Q2_K: return max_compute_capability >= CC_RDNA2 ? 128 : 32; case GGML_TYPE_Q3_K: @@ -6273,6 +6275,8 @@ static int64_t get_row_rounding(ggml_type type) { return 64; case GGML_TYPE_F16: return 1; + case GGML_TYPE_F32: + return 1; case GGML_TYPE_Q2_K: case GGML_TYPE_Q3_K: case GGML_TYPE_Q4_K: From bf46304cbc0d19b4059a224559748c816771170a Mon Sep 17 00:00:00 2001 From: Andrew Godfrey Date: Wed, 15 Nov 2023 18:58:05 -0800 Subject: [PATCH 2/3] Update ggml-cuda.cu Co-authored-by: Jared Van Bortel --- ggml-cuda.cu | 1 - 1 file changed, 1 deletion(-) diff --git a/ggml-cuda.cu b/ggml-cuda.cu index 58e7ecb2ed40d..6a29795a9562d 100644 --- a/ggml-cuda.cu +++ b/ggml-cuda.cu @@ -6250,7 +6250,6 @@ static int64_t get_row_rounding(ggml_type type) { case GGML_TYPE_Q8_0: return max_compute_capability >= CC_RDNA2 ? 128 : 64; case GGML_TYPE_F16: - return 1; case GGML_TYPE_F32: return 1; case GGML_TYPE_Q2_K: From a55a095119a57d847aeb239afc1f8b328d8eefcd Mon Sep 17 00:00:00 2001 From: Andrew Godfrey Date: Wed, 15 Nov 2023 18:58:12 -0800 Subject: [PATCH 3/3] Update ggml-cuda.cu Co-authored-by: Jared Van Bortel --- ggml-cuda.cu | 1 - 1 file changed, 1 deletion(-) diff --git a/ggml-cuda.cu b/ggml-cuda.cu index 6a29795a9562d..d2fb377d9c289 100644 --- a/ggml-cuda.cu +++ b/ggml-cuda.cu @@ -6273,7 +6273,6 @@ static int64_t get_row_rounding(ggml_type type) { case GGML_TYPE_Q8_0: return 64; case GGML_TYPE_F16: - return 1; case GGML_TYPE_F32: return 1; case GGML_TYPE_Q2_K: