vrr/koboldcpp
2
0
Fork 0
mirror of https://github.com/LostRuins/koboldcpp.git synced 2025-09-11 17:44:38 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions 3

try to fix loras

Browse source
This commit is contained in:
Concedo 2024-05-15 16:05:51 +08:00
parent ebbb969526
commit e1e6833a7a
2 changed files with 23 additions and 2 deletions
Download patch file Download diff file Expand all files Collapse all files

4
otherarch/sdcpp/sdtype_adapter.cpp
View file

@ -251,8 +251,8 @@ bool sdtype_load_model(const sd_load_model_inputs inputs) {
if(lorafilename!="" && inputs.lora_multiplier>0) if(lorafilename!="" && inputs.lora_multiplier>0)
{ {
printf("\nApplying LoRA now...\n"); printf("\nSet pending LoRA...\n");
sd_ctx->sd->apply_lora_from_file(lorafilename,inputs.lora_multiplier); sd_ctx->sd->set_pending_lora(lorafilename,inputs.lora_multiplier);
} }
return true; return true;

21
otherarch/sdcpp/stable-diffusion.cpp
View file

@ -57,6 +57,9 @@ void calculate_alphas_cumprod(float* alphas_cumprod,
/*=============================================== StableDiffusionGGML ================================================*/ /*=============================================== StableDiffusionGGML ================================================*/
static std::string pending_apply_lora_fname = "";
static float pending_apply_lora_power = 1.0f;
class StableDiffusionGGML { class StableDiffusionGGML {
public: public:
ggml_backend_t backend = NULL; // general backend ggml_backend_t backend = NULL; // general backend
@ -87,6 +90,7 @@ public:
// lora_name => multiplier // lora_name => multiplier
std::unordered_map<std::string, float> curr_lora_state; std::unordered_map<std::string, float> curr_lora_state;
std::shared_ptr<Denoiser> denoiser = std::make_shared<CompVisDenoiser>(); std::shared_ptr<Denoiser> denoiser = std::make_shared<CompVisDenoiser>();
StableDiffusionGGML() = default; StableDiffusionGGML() = default;
@ -402,6 +406,11 @@ public:
return result < -1; return result < -1;
} }
void set_pending_lora(const std::string& lora_path, float multiplier) {
pending_apply_lora_fname = lora_path;
pending_apply_lora_power = multiplier;
}
void apply_lora_from_file(const std::string& lora_path, float multiplier) { void apply_lora_from_file(const std::string& lora_path, float multiplier) {
int64_t t0 = ggml_time_ms(); int64_t t0 = ggml_time_ms();
std::string st_file_path = lora_path; std::string st_file_path = lora_path;
@ -1394,6 +1403,12 @@ sd_image_t* txt2img(sd_ctx_t* sd_ctx,
int64_t t0 = ggml_time_ms(); int64_t t0 = ggml_time_ms();
sd_ctx->sd->apply_loras(lora_f2m); sd_ctx->sd->apply_loras(lora_f2m);
if(pending_apply_lora_fname!="" && pending_apply_lora_power>0)
{
printf("\nApplying LoRA now...\n");
sd_ctx->sd->apply_lora_from_file(pending_apply_lora_fname,pending_apply_lora_power);
pending_apply_lora_fname = "";
}
int64_t t1 = ggml_time_ms(); int64_t t1 = ggml_time_ms();
LOG_INFO("apply_loras completed, taking %.2fs", (t1 - t0) * 1.0f / 1000); LOG_INFO("apply_loras completed, taking %.2fs", (t1 - t0) * 1.0f / 1000);
struct ggml_init_params params; struct ggml_init_params params;
@ -1587,6 +1602,12 @@ sd_image_t* img2img(sd_ctx_t* sd_ctx,
// load lora from file // load lora from file
int64_t t0 = ggml_time_ms(); int64_t t0 = ggml_time_ms();
sd_ctx->sd->apply_loras(lora_f2m); sd_ctx->sd->apply_loras(lora_f2m);
if(pending_apply_lora_fname!="" && pending_apply_lora_power>0)
{
printf("\nApplying LoRA now...\n");
sd_ctx->sd->apply_lora_from_file(pending_apply_lora_fname,pending_apply_lora_power);
pending_apply_lora_fname = "";
}
int64_t t1 = ggml_time_ms(); int64_t t1 = ggml_time_ms();
LOG_INFO("apply_loras completed, taking %.2fs", (t1 - t0) * 1.0f / 1000); LOG_INFO("apply_loras completed, taking %.2fs", (t1 - t0) * 1.0f / 1000);