vrr/koboldcpp
2
0
Fork 0
mirror of https://github.com/LostRuins/koboldcpp.git synced 2025-09-11 09:34:37 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

fixed incorrect auto gpu settings, fixed clblast not working

Browse source
This commit is contained in:
Concedo 2024-09-21 17:59:52 +08:00
parent 004a35b16d
commit 229108f877
2 changed files with 9 additions and 3 deletions
Download patch file Download diff file Expand all files Collapse all files

8
ggml/src/ggml-backend.c
View file

@ -1174,6 +1174,8 @@ static char causes[GGML_DEFAULT_GRAPH_SIZE*16 + GGML_SCHED_MAX_SPLITS_DEBUG*GGML
#define GET_CAUSE(node) "" #define GET_CAUSE(node) ""
#endif #endif
static bool backend_prealloc_warn = false;
// returns the backend that should be used for the node based on the current locations // returns the backend that should be used for the node based on the current locations
static int ggml_backend_sched_backend_id_from_cur(ggml_backend_sched_t sched, struct ggml_tensor * tensor) { static int ggml_backend_sched_backend_id_from_cur(ggml_backend_sched_t sched, struct ggml_tensor * tensor) {
// TODO: use supports_op to check if the backend supports the op // TODO: use supports_op to check if the backend supports the op
@ -1196,7 +1198,11 @@ static int ggml_backend_sched_backend_id_from_cur(ggml_backend_sched_t sched, st
if (tensor->buffer || (tensor->view_src && tensor->view_src->buffer)) { if (tensor->buffer || (tensor->view_src && tensor->view_src->buffer)) {
// since the tensor is pre-allocated, it cannot be moved to another backend // since the tensor is pre-allocated, it cannot be moved to another backend
GGML_ABORT("pre-allocated tensor in a backend that cannot run the operation"); if(!backend_prealloc_warn)
{
backend_prealloc_warn = true;
printf("\nCaution: pre-allocated tensor in a backend that cannot run the operation\n");
}
} }
// graph input // graph input

4
koboldcpp.py
View file

@ -41,7 +41,7 @@ maxhordelen = 400
modelbusy = threading.Lock() modelbusy = threading.Lock()
requestsinqueue = 0 requestsinqueue = 0
defaultport = 5001 defaultport = 5001
KcppVersion = "1.75" KcppVersion = "1.75.1"
showdebug = True showdebug = True
guimode = False guimode = False
showsamplerwarning = True showsamplerwarning = True
@ -4038,7 +4038,7 @@ def main(launch_args,start_server=True):
print(f"MacOS detected: Auto GPU layers set to maximum") print(f"MacOS detected: Auto GPU layers set to maximum")
args.gpulayers = 200 args.gpulayers = 200
elif not shouldavoidgpu and args.model_param and os.path.exists(args.model_param): elif not shouldavoidgpu and args.model_param and os.path.exists(args.model_param):
if not args.usecublas and (args.usevulkan is None) and not args.useclblast: if (args.usecublas is None) and (args.usevulkan is None) and (args.useclblast is None):
print("No GPU or CPU backend was selected. Trying to assign one for you automatically...") print("No GPU or CPU backend was selected. Trying to assign one for you automatically...")
auto_set_backend_cli() auto_set_backend_cli()
if MaxMemory[0] == 0: #try to get gpu vram for cuda if not picked yet if MaxMemory[0] == 0: #try to get gpu vram for cuda if not picked yet