vbatts/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

cuda : fix array size + indents

Browse source 
ggml-ci
This commit is contained in:
Georgi Gerganov 2024-06-02 19:58:50 +03:00
parent 61e0a84fff
commit ddac1ef681
No known key found for this signature in database
GPG key ID: 449E073F9DC10735
2 changed files with 3 additions and 3 deletions
Download patch file Download diff file Expand all files Collapse all files

2
ggml-cuda/rope.cu
View file

@ -1,7 +1,7 @@
#include "rope.cuh" #include "rope.cuh"
struct rope_corr_dims { struct rope_corr_dims {
float v[4]; // TODO: is there any reson for this to be 4 instead of 2? float v[2];
}; };
static __device__ float rope_yarn_ramp(const float low, const float high, const int i0) { static __device__ float rope_yarn_ramp(const float low, const float high, const int i0) {

4
ggml.c
View file

@ -14371,7 +14371,7 @@ static void ggml_compute_forward_rope_f32(
const float sin_theta = cache[i0 + 1]; const float sin_theta = cache[i0 + 1];
const float * const src = (float *)((char *) src0->data + i3*nb03 + i2*nb02 + i1*nb01 + i0*nb00); const float * const src = (float *)((char *) src0->data + i3*nb03 + i2*nb02 + i1*nb01 + i0*nb00);
float * dst_data = (float *)((char *) dst->data + i3*nb3 + i2*nb2 + i1*nb1 + i0*nb0); float * dst_data = (float *)((char *) dst->data + i3*nb3 + i2*nb2 + i1*nb1 + i0*nb0);
const float x0 = src[0]; const float x0 = src[0];
const float x1 = src[1]; const float x1 = src[1];
@ -14500,7 +14500,7 @@ static void ggml_compute_forward_rope_f16(
const float sin_theta = cache[i0 + 1]; const float sin_theta = cache[i0 + 1];
const ggml_fp16_t * const src = (ggml_fp16_t *)((char *) src0->data + i3*nb03 + i2*nb02 + i1*nb01 + i0*nb00); const ggml_fp16_t * const src = (ggml_fp16_t *)((char *) src0->data + i3*nb03 + i2*nb02 + i1*nb01 + i0*nb00);
ggml_fp16_t * dst_data = (ggml_fp16_t *)((char *) dst->data + i3*nb3 + i2*nb2 + i1*nb1 + i0*nb0); ggml_fp16_t * dst_data = (ggml_fp16_t *)((char *) dst->data + i3*nb3 + i2*nb2 + i1*nb1 + i0*nb0);
const float x0 = GGML_FP16_TO_FP32(src[0]); const float x0 = GGML_FP16_TO_FP32(src[0]);
const float x1 = GGML_FP16_TO_FP32(src[1]); const float x1 = GGML_FP16_TO_FP32(src[1]);