vrr/koboldcpp
2
0
Fork 0
mirror of https://github.com/LostRuins/koboldcpp.git synced 2026-05-20 09:25:53 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions 3

Merge branch 'concedo' into concedo_experimental

Browse source
This commit is contained in:
Concedo 2026-01-05 15:24:43 +08:00
commit 301a04adfc
3 changed files with 6 additions and 7 deletions
Download patch file Download diff file Expand all files Collapse all files

1
ggml/src/ggml-cuda/common.cuh
View file

@ -1065,7 +1065,6 @@ struct ggml_cuda_graph {
cudaGraphExec_t instance = nullptr;
size_t num_nodes = 0;
std::vector<cudaGraphNode_t> nodes;
std::vector<cudaKernelNodeParams> params;
bool disable_due_to_gpu_arch = false;
bool disable_due_to_too_many_updates = false;
bool disable_due_to_failed_graph_capture = false;

10
ggml/src/ggml-cuda/ggml-cuda.cu
View file

@ -3708,6 +3708,7 @@ static void evaluate_and_capture_cuda_graph(ggml_backend_cuda_context * cuda_ctx
}
static bool ggml_cuda_set_cuda_graph_enabled(ggml_backend_cuda_context * cuda_ctx) {
#ifdef USE_CUDA_GRAPH
static const bool disable_cuda_graphs_due_to_env = (getenv("GGML_CUDA_DISABLE_GRAPHS") != nullptr);
@ -3748,17 +3749,15 @@ static bool ggml_cuda_set_cuda_graph_enabled(ggml_backend_cuda_context * cuda_ct
static enum ggml_status ggml_backend_cuda_graph_compute(ggml_backend_t backend, ggml_cgraph * cgraph) {
ggml_backend_cuda_context * cuda_ctx = (ggml_backend_cuda_context *) backend->context;
ggml_cuda_set_device(cuda_ctx->device);
bool use_cuda_graph = false;
bool cuda_graph_update_required = false;
// graph_optimize calls set_cuda_graph_enabled, in-case it not called (i.e. graph_compute is directly called)
// we call it here instead.
#ifdef USE_CUDA_GRAPH
if (!cuda_ctx->cuda_graph) {
use_cuda_graph = ggml_cuda_set_cuda_graph_enabled(cuda_ctx);
} else {
use_cuda_graph = cuda_ctx->cuda_graph && cuda_ctx->cuda_graph->cuda_graphs_enabled;
}
use_cuda_graph = ggml_cuda_set_cuda_graph_enabled(cuda_ctx);
if (use_cuda_graph) {
cuda_graph_update_required = is_cuda_graph_update_required(cuda_ctx, cgraph);
@ -3774,6 +3773,7 @@ static enum ggml_status ggml_backend_cuda_graph_compute(ggml_backend_t backend,
if (cuda_ctx->cuda_graph->number_consecutive_updates >= 4) {
cuda_ctx->cuda_graph->disable_due_to_too_many_updates = true;
cuda_ctx->cuda_graph->cuda_graphs_enabled = false;
#ifndef NDEBUG
GGML_LOG_DEBUG("%s: disabling CUDA graphs due to too many consecutive updates\n", __func__);
#endif

2
koboldcpp.py
View file

@ -67,7 +67,7 @@ dry_seq_break_max = 128
extra_images_max = 4 # for kontext/qwen img
# global vars
KcppVersion = "1.105.2"
KcppVersion = "1.105.3"
showdebug = True
kcpp_instance = None #global running instance
global_memory = {"tunnel_url": "", "restart_target":"", "input_to_exit":False, "load_complete":False, "restart_override_config_target":""}