Files
llama.cpp/ggml-cuda/convert.cuh
T

14 lines
391 B
Plaintext
Raw Normal View History

2024-03-25 13:50:23 +01:00
#include "common.cuh"
#define CUDA_DEQUANTIZE_BLOCK_SIZE 256
template<typename T>
2024-04-09 09:16:13 +01:00
using to_t_cuda_t = void (*)(const void * __restrict__ x, T * __restrict__ y, int64_t k, cudaStream_t stream);
2024-03-25 13:50:23 +01:00
typedef to_t_cuda_t<float> to_fp32_cuda_t;
typedef to_t_cuda_t<half> to_fp16_cuda_t;
to_fp16_cuda_t ggml_get_to_fp16_cuda(ggml_type type);
to_fp32_cuda_t ggml_get_to_fp32_cuda(ggml_type type);