matt/ollama37
1
0
Fork 0
mirror of https://github.com/dogkeeper886/ollama37.git synced 2025-12-10 15:57:04 +00:00
Code Issues Packages Projects Releases Wiki Activity

Refine build to support CPU only

Browse Source 
If someone checks out the ollama repo and doesn't install the CUDA
library, this will ensure they can build a CPU only version
This commit is contained in:
Daniel Hiltgen
2023-12-13 17:26:47 -08:00
parent 51082535e1
commit 1b991d0ba9
9 changed files with 152 additions and 98 deletions
Download Patch File Download Diff File Expand all files Collapse all files

24
gpu/gpu_info_cuda.c
View File

@@ -19,6 +19,7 @@ const char *cuda_lib_paths[] = {
#endif
void cuda_init(cuda_init_resp_t *resp) {
nvmlReturn_t ret;
resp->err = NULL;
const int buflen = 256;
char buf[buflen + 1];
@@ -56,6 +57,13 @@ void cuda_init(cuda_init_resp_t *resp) {
return;
}
}
ret = (*resp->ch.initFn)();
if (ret != NVML_SUCCESS) {
snprintf(buf, buflen, "nvml vram init failure: %d", ret);
resp->err = strdup(buf);
}
return;
}
@@ -73,17 +81,9 @@ void cuda_check_vram(cuda_handle_t h, mem_info_t *resp) {
return;
}
ret = (*h.initFn)();
if (ret != NVML_SUCCESS) {
snprintf(buf, buflen, "nvml vram init failure: %d", ret);
resp->err = strdup(buf);
return;
}
// TODO - handle multiple GPUs
ret = (*h.getHandle)(0, &device);
if (ret != NVML_SUCCESS) {
(*h.shutdownFn)();
snprintf(buf, buflen, "unable to get device handle: %d", ret);
resp->err = strdup(buf);
return;
@@ -91,20 +91,12 @@ void cuda_check_vram(cuda_handle_t h, mem_info_t *resp) {
ret = (*h.getMemInfo)(device, &memInfo);
if (ret != NVML_SUCCESS) {
(*h.shutdownFn)();
snprintf(buf, buflen, "device memory info lookup failure: %d", ret);
resp->err = strdup(buf);
return;
}
resp->total = memInfo.total;
resp->free = memInfo.free;
ret = (*h.shutdownFn)();
if (ret != NVML_SUCCESS) {
snprintf(buf, buflen, "nvml vram shutdown failure: %d", ret);
resp->err = strdup(buf);
}
return;
}
#endif // __APPLE__