upstream-archive/stable-diffusion.cpp
1
0
Fork 0
mirror of https://github.com/leejet/stable-diffusion.cpp.git synced 2026-07-23 11:20:53 -05:00
Code Issues Packages Projects Releases Wiki Activity

feat: add TAESD implementation - faster autoencoder (#88)

Browse Source 
* add taesd implementation

* taesd gpu offloading

* show seed when generating image with -s -1

* less restrictive with larger images

* cuda: im2col speedup x2

* cuda: group norm speedup x90

* quantized models now works in cuda :)

* fix cal mem size

---------

Co-authored-by: leejet <leejet714@gmail.com>
This commit is contained in:
Steward Garcia
2023-12-05 09:40:03 -05:00
committed by GitHub
parent f99bcd1f76
commit 134883aec4
14 changed files with 908 additions and 46904 deletions
Download Patch File Download Diff File Expand all files Collapse all files

19
model.cpp
View File

@@ -1296,7 +1296,7 @@ bool ModelLoader::load_tensors(on_new_tensor_cb_t on_new_tensor_cb) {
if (backend == NULL || ggml_backend_is_cpu(backend)) {
// for the CPU and Metal backend, we can copy directly into the tensor
if (tensor_storage.type == dst_tensor->type) {
GGML_ASSERT(ggml_nbytes(dst_tensor) == nbytes_to_read);
GGML_ASSERT(ggml_nbytes(dst_tensor) == tensor_storage.nbytes());
read_data(tensor_storage, (char*)dst_tensor->data, nbytes_to_read);
if (tensor_storage.is_bf16) {
@@ -1349,16 +1349,23 @@ bool ModelLoader::load_tensors(on_new_tensor_cb_t on_new_tensor_cb) {
return success;
}
int64_t ModelLoader::cal_mem_size() {
int64_t ModelLoader::cal_mem_size(ggml_backend_t backend) {
size_t alignment = 128;
if (backend != NULL) {
alignment = ggml_backend_get_alignment(backend);
}
int64_t mem_size = 0;
std::vector<TensorStorage> processed_tensor_storages;
for (auto& tensor_storage : tensor_storages) {
if (is_unused_tensor(tensor_storage.name)) {
continue;
}
mem_size += tensor_storage.nbytes();
mem_size += GGML_MEM_ALIGN * 2; // for lora alphas
preprocess_tensor(tensor_storage, processed_tensor_storages);
}
return mem_size + 10 * 1024 * 1024;
for (auto& tensor_storage : processed_tensor_storages) {
mem_size += tensor_storage.nbytes() + alignment;
}
return mem_size;
}