vrr/koboldcpp
2
0
Fork 0
mirror of https://github.com/LostRuins/koboldcpp.git synced 2026-07-10 01:18:32 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

testing max nodes per submit

Browse source
This commit is contained in:
Concedo 2026-06-30 22:26:02 +08:00
parent 61ad97cbc1
commit ca3a77a87b
2 changed files with 2 additions and 2 deletions
Download patch file Download diff file Expand all files Collapse all files

2
ggml/src/ggml-vulkan/ggml-vulkan.cpp
View file

@ -5908,7 +5908,7 @@ static vk_device ggml_vk_get_device(size_t idx) {
// Submit at least every 100 nodes, in case there are workloads without as much matmul.
// device->max_nodes_per_submit = 100;
device->max_nodes_per_submit = device->uma ? 8 : 64; //kcpp fix for https://github.com/ggml-org/llama.cpp/issues/21724
device->max_nodes_per_submit = device->uma ? 4 : 16; //kcpp fix for https://github.com/ggml-org/llama.cpp/issues/21724
const char* GGML_VK_MAX_NODES_PER_SUBMIT = getenv("GGML_VK_MAX_NODES_PER_SUBMIT");
if (GGML_VK_MAX_NODES_PER_SUBMIT != nullptr) {
uint32_t max_nodes_per_submit = std::stoul(GGML_VK_MAX_NODES_PER_SUBMIT);

2
koboldcpp.py
View file

@ -12225,7 +12225,7 @@ if __name__ == '__main__':
modelgroup.add_argument("--model","-m", metavar=('[filenames]'), help="Model file to load. Accepts multiple values if they are URLs.", type=str, nargs='+', default=[])
modelgroup.add_argument("model_param", help="Model file to load (positional)", nargs="?")
parser.add_argument("--config", metavar=('[filename]'), help="Load settings from a .kcpps file. Other arguments will be ignored", type=str, nargs=1)
parser.add_argument("--contextsize","--ctx-size", "-c", help=f"Controls the memory allocated for maximum context size, only change if you need more RAM for big contexts. (default {default_maxctx}).",metavar=('[256 to 262144]'), type=check_range(int,256,262144), default=default_maxctx)
parser.add_argument("--contextsize","--ctx-size", "-c", help=f"Controls the memory allocated for maximum context size, only change if you need more RAM for big contexts. (default {default_maxctx}).",metavar=('[256 to 524288]'), type=check_range(int,256,524288), default=default_maxctx)
parser.add_argument("--gpulayers","--gpu-layers","--n-gpu-layers","-ngl", help="Set number of layers to offload to GPU (when using GPU). Set to -1 to enable autofit (default), set to 0 to disable GPU offload.",metavar=('[GPU layers]'), nargs='?', const=1, type=int, default=-1)
parser.add_argument("--host", metavar=('[ipaddr]'), help="Host IP to listen on. If this flag is not set, all routable interfaces are accepted.", default="")
parser.add_argument("--launch", help="Launches a web browser when load is completed.", action='store_true')