1 files changed, 0 insertions, 20 deletions
diff --git a/ggml-cuda/quantize.cuh b/ggml-cuda/quantize.cuh
deleted file mode 100644
index 486c9360..00000000
--- a/ggml-cuda/quantize.cuh
+++ /dev/null
@@ -1,20 +0,0 @@
-#pragma once
-
-#include "common.cuh"
-#include "mmq.cuh"
-
-#include <cstdint>
-
-#define CUDA_QUANTIZE_BLOCK_SIZE 256
-
-typedef void (*quantize_cuda_t)(
-    const float * x, void * vy, const int64_t kx0, const int64_t kx1, const int64_t channels, const int64_t kx0_padded,
-    const ggml_type type_x, cudaStream_t stream);
-
-void quantize_row_q8_1_cuda(
-    const float * x, void * vy, const int64_t kx0, const int64_t kx1, const int64_t channels, const int64_t kx0_padded,
-    const ggml_type type_x, cudaStream_t stream);
-
-void quantize_mmq_q8_1_cuda(
-    const float * x, void * vy, const int64_t kx0, const int64_t kx1, const int64_t channels, const int64_t kx0_padded,
-    const ggml_type type_x, cudaStream_t stream);