upstream-archive/stable-diffusion.cpp
1
0
Fork 0
mirror of https://github.com/leejet/stable-diffusion.cpp.git synced 2026-07-31 15:20:41 -05:00
Code Issues Packages Projects Releases Wiki Activity

feat: add max-vram based segmented param offload (#1476)

Browse Source
This commit is contained in:
leejet
2026-05-06 21:56:02 +08:00
committed by GitHub
parent 586b6f1481
commit 90e87bc846
26 changed files with 1576 additions and 66 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
src/upscaler.cpp
View File

@@ -12,6 +12,13 @@ UpscalerGGML::UpscalerGGML(int n_threads,
tile_size(tile_size) {
}
void UpscalerGGML::set_max_graph_vram_bytes(size_t max_vram_bytes) {
max_graph_vram_bytes = max_vram_bytes;
if (esrgan_upscaler) {
esrgan_upscaler->set_max_graph_vram_bytes(max_vram_bytes);
}
}
bool UpscalerGGML::load_from_file(const std::string& esrgan_path,
bool offload_params_to_cpu,
int n_threads) {
@@ -30,6 +37,7 @@ bool UpscalerGGML::load_from_file(const std::string& esrgan_path,
}
LOG_INFO("Upscaler weight type: %s", ggml_type_name(model_data_type));
esrgan_upscaler = std::make_shared<ESRGAN>(backend, offload_params_to_cpu, tile_size, model_loader.get_tensor_storage_map());
esrgan_upscaler->set_max_graph_vram_bytes(max_graph_vram_bytes);
if (direct) {
esrgan_upscaler->set_conv2d_direct_enabled(true);
}