vrr/koboldcpp
2
0
Fork 0
mirror of https://github.com/LostRuins/koboldcpp.git synced 2026-08-22 06:45:11 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

fix for https://github.com/ggml-org/llama.cpp/issues/21724 , and use wbruna's heuristic https://github.com/ggml-org/llama.cpp/pull/24872

Browse source
This commit is contained in:
Concedo 2026-06-27 16:44:48 +08:00
parent 16f197ab86
commit 1783236b05
1 changed files with 2 additions and 1 deletions
Download patch file Download diff file Expand all files Collapse all files

3
ggml/src/ggml-vulkan/ggml-vulkan.cpp
View file

@ -5907,7 +5907,8 @@ static vk_device ggml_vk_get_device(size_t idx) {
(vk11_props.subgroupSupportedOperations & vk::SubgroupFeatureFlagBits::eVote);
// 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 = 100;
device->max_nodes_per_submit = device->uma ? 8 : 64; //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);