From a48e16324770bb829406d06e11be1df0c8a3b517 Mon Sep 17 00:00:00 2001
From: Kawrakow <iwankawrakow@gmail.com>
Date: Mon, 10 Mar 2025 16:19:09 +0200
Subject: DeepSeek imatrix stuff (#250)

* This gives us ~20% TG speedup for DeepSeek on CUDA

* Slightly better

* Also do it for plain (not fused) mul_mat_id

* Guard against numerical precision issues for MLA on CUDA

* imatrix: wv_b <-> wkv_b

---------

Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
---
 ggml/src/ggml-cuda/cpy.cu | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

(limited to 'ggml/src/ggml-cuda')

diff --git a/ggml/src/ggml-cuda/cpy.cu b/ggml/src/ggml-cuda/cpy.cu
index 0b269a86..fabe8843 100644
--- a/ggml/src/ggml-cuda/cpy.cu
+++ b/ggml/src/ggml-cuda/cpy.cu
@@ -556,7 +556,7 @@ void* ggml_cuda_cpy_fn(const ggml_tensor * src0, ggml_tensor * src1) {
     } else if (src0->type == GGML_TYPE_F32 && src1->type == GGML_TYPE_Q6_0) {
             return (void*) cpy_f32_q<cpy_blck_f32_q6_0, QK6_0>;
     } else if (src0->type == GGML_TYPE_F16 && src1->type == GGML_TYPE_F16) {
-            return (void*) cpy_f32_f16<cpy_1_f32_f16>;
+            return (void*) cpy_f32_f16<cpy_1_f16_f16>;
     } else if (src0->type == GGML_TYPE_F16 && src1->type == GGML_TYPE_F32) {
             return (void*) cpy_f32_f16<cpy_1_f16_f32>;
     } else {
-- 
cgit v1.2.3