vrr/prima.cpp
2
0
Fork 0
mirror of https://github.com/Lizonghang/prima.cpp.git synced 2025-09-08 10:09:03 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

fix cuda compatibility errors

Browse source
This commit is contained in:
Lizonghang 2024-11-26 22:35:58 +04:00
parent 3f008f2ad9
commit 0a91ad3edc
2 changed files with 7 additions and 5 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

10
common/profiler.cpp
View file

@ -21,6 +21,7 @@
#ifdef GGML_USE_CUDA #ifdef GGML_USE_CUDA
#include "ggml-cuda.h" #include "ggml-cuda.h"
#include <cuda_runtime.h>
#endif #endif
#include <cmath> #include <cmath>
@ -33,7 +34,6 @@
#include <vector> #include <vector>
#include <inttypes.h> #include <inttypes.h>
#include <thread> #include <thread>
#include <cuda_runtime.h>
const char * device_name() { const char * device_name() {
static char device_name[256]; static char device_name[256];
@ -522,10 +522,7 @@ float device_memory_bw(int n_thread) {
} }
float device_cuda_memory_bw(struct llama_model * model) { float device_cuda_memory_bw(struct llama_model * model) {
#ifndef GGML_USE_CUDA #ifdef GGML_USE_CUDA
return 0.0f;
#endif
const int n_embd = llama_n_embd(model) * 2; const int n_embd = llama_n_embd(model) * 2;
std::vector<float> matrix_A(n_embd * n_embd, 1.0f); std::vector<float> matrix_A(n_embd * n_embd, 1.0f);
@ -581,6 +578,9 @@ float device_cuda_memory_bw(struct llama_model * model) {
ggml_backend_free(backend); ggml_backend_free(backend);
return bandwidth; return bandwidth;
#else
return 0.0f;
#endif
} }
int device_has_metal(void) { int device_has_metal(void) {

2
ggml/src/ggml.c
View file

@ -17564,6 +17564,7 @@ static void ggml_compute_forward(struct ggml_compute_params * params, struct ggm
ggml_compute_forward_opt_step_adamw(params, tensor); ggml_compute_forward_opt_step_adamw(params, tensor);
} }
break; break;
case GGML_OP_READ:
case GGML_OP_NONE: case GGML_OP_NONE:
{ {
// nop // nop
@ -18719,6 +18720,7 @@ static void ggml_compute_backward(struct ggml_context * ctx, struct ggml_tensor
{ {
GGML_ABORT("fatal error"); // not supported GGML_ABORT("fatal error"); // not supported
} }
case GGML_OP_READ:
case GGML_OP_NONE: case GGML_OP_NONE:
{ {
// nop // nop