upstream-archive/stable-diffusion.cpp
1
0
Fork 0
mirror of https://github.com/leejet/stable-diffusion.cpp.git synced 2026-07-26 04:40:51 -05:00
Code Issues Packages Projects Releases Wiki Activity

refactor: add some sd vesion helper functions

Browse Source
This commit is contained in:
leejet
2024-11-23 13:02:44 +08:00
parent 1c168d98a5
commit b5f4932696
5 changed files with 59 additions and 38 deletions
Download Patch File Download Diff File Expand all files Collapse all files

46
stable-diffusion.cpp
View File

@@ -286,9 +286,9 @@ public:
"try specifying SDXL VAE FP16 Fix with the --vae parameter. "
"You can find it here: https://huggingface.co/madebyollin/sdxl-vae-fp16-fix/blob/main/sdxl_vae.safetensors");
}
} else if (version == VERSION_SD3_2B || version == VERSION_SD3_5_8B || version == VERSION_SD3_5_2B) {
} else if (sd_version_is_sd3(version)) {
scale_factor = 1.5305f;
} else if (version == VERSION_FLUX_DEV || version == VERSION_FLUX_SCHNELL || version == VERSION_FLUX_LITE) {
} else if (sd_version_is_flux(version)) {
scale_factor = 0.3611;
// TODO: shift_factor
}
@@ -309,7 +309,7 @@ public:
} else {
clip_backend = backend;
bool use_t5xxl = false;
if (version == VERSION_SD3_2B || version == VERSION_SD3_5_8B || version == VERSION_SD3_5_2B || version == VERSION_FLUX_DEV || version == VERSION_FLUX_SCHNELL || version == VERSION_FLUX_LITE) {
if (sd_version_is_dit(version)) {
use_t5xxl = true;
}
if (!ggml_backend_is_cpu(backend) && use_t5xxl && conditioner_wtype != GGML_TYPE_F32) {
@@ -323,18 +323,18 @@ public:
if (diffusion_flash_attn) {
LOG_INFO("Using flash attention in the diffusion model");
}
if (version == VERSION_SD3_2B || version == VERSION_SD3_5_8B || version == VERSION_SD3_5_2B) {
if (sd_version_is_sd3(version)) {
if (diffusion_flash_attn) {
LOG_WARN("flash attention in this diffusion model is currently unsupported!");
}
cond_stage_model = std::make_shared<SD3CLIPEmbedder>(clip_backend, conditioner_wtype);
diffusion_model = std::make_shared<MMDiTModel>(backend, diffusion_model_wtype, version);
} else if (version == VERSION_FLUX_DEV || version == VERSION_FLUX_SCHNELL || version == VERSION_FLUX_LITE) {
} else if (sd_version_is_flux(version)) {
cond_stage_model = std::make_shared<FluxCLIPEmbedder>(clip_backend, conditioner_wtype);
diffusion_model = std::make_shared<FluxModel>(backend, diffusion_model_wtype, version, diffusion_flash_attn);
} else {
if (id_embeddings_path.find("v2") != std::string::npos) {
cond_stage_model = std::make_shared<FrozenCLIPEmbedderWithCustomWords>(clip_backend, conditioner_wtype, embeddings_path, version, VERSION_2);
cond_stage_model = std::make_shared<FrozenCLIPEmbedderWithCustomWords>(clip_backend, conditioner_wtype, embeddings_path, version, PM_VERSION_2);
} else {
cond_stage_model = std::make_shared<FrozenCLIPEmbedderWithCustomWords>(clip_backend, conditioner_wtype, embeddings_path, version);
}
@@ -373,7 +373,7 @@ public:
}
if (id_embeddings_path.find("v2") != std::string::npos) {
pmid_model = std::make_shared<PhotoMakerIDEncoder>(backend, model_wtype, version, VERSION_2);
pmid_model = std::make_shared<PhotoMakerIDEncoder>(backend, model_wtype, version, PM_VERSION_2);
LOG_INFO("using PhotoMaker Version 2");
} else {
pmid_model = std::make_shared<PhotoMakerIDEncoder>(backend, model_wtype, version);
@@ -527,10 +527,10 @@ public:
is_using_v_parameterization = true;
}
if (version == VERSION_SD3_2B || version == VERSION_SD3_5_8B || version == VERSION_SD3_5_2B) {
if (sd_version_is_sd3(version)) {
LOG_INFO("running in FLOW mode");
denoiser = std::make_shared<DiscreteFlowDenoiser>();
} else if (version == VERSION_FLUX_DEV || version == VERSION_FLUX_SCHNELL || version == VERSION_FLUX_LITE) {
} else if (sd_version_is_flux(version)) {
LOG_INFO("running in Flux FLOW mode");
float shift = 1.15f;
if (version == VERSION_FLUX_SCHNELL) {
@@ -804,7 +804,7 @@ public:
out_uncond = ggml_dup_tensor(work_ctx, x);
}
if (has_skiplayer) {
if (version == VERSION_SD3_2B || version == VERSION_SD3_5_2B || version == VERSION_SD3_5_8B || version == VERSION_FLUX_DEV || version == VERSION_FLUX_SCHNELL) {
if (sd_version_is_dit(version)) {
out_skip = ggml_dup_tensor(work_ctx, x);
} else {
has_skiplayer = false;
@@ -995,9 +995,9 @@ public:
if (use_tiny_autoencoder) {
C = 4;
} else {
if (version == VERSION_SD3_2B || version == VERSION_SD3_5_8B || version == VERSION_SD3_5_2B) {
if (sd_version_is_sd3(version)) {
C = 32;
} else if (version == VERSION_FLUX_DEV || version == VERSION_FLUX_SCHNELL || version == VERSION_FLUX_LITE) {
} else if (sd_version_is_flux(version)) {
C = 32;
}
}
@@ -1214,7 +1214,7 @@ sd_image_t* generate_image(sd_ctx_t* sd_ctx,
}
// preprocess input id images
std::vector<sd_image_t*> input_id_images;
bool pmv2 = sd_ctx->sd->pmid_model->get_version() == VERSION_2;
bool pmv2 = sd_ctx->sd->pmid_model->get_version() == PM_VERSION_2;
if (sd_ctx->sd->pmid_model && input_id_images_path.size() > 0) {
std::vector<std::string> img_files = get_files_from_dir(input_id_images_path);
for (std::string img_file : img_files) {
@@ -1343,9 +1343,9 @@ sd_image_t* generate_image(sd_ctx_t* sd_ctx,
// Sample
std::vector<struct ggml_tensor*> final_latents; // collect latents to decode
int C = 4;
if (sd_ctx->sd->version == VERSION_SD3_2B || sd_ctx->sd->version == VERSION_SD3_5_8B || sd_ctx->sd->version == VERSION_SD3_5_2B) {
if (sd_version_is_sd3(sd_ctx->sd->version)) {
C = 16;
} else if (sd_ctx->sd->version == VERSION_FLUX_DEV || sd_ctx->sd->version == VERSION_FLUX_SCHNELL || sd_ctx->sd->version == VERSION_FLUX_LITE) {
} else if (sd_version_is_flux(sd_ctx->sd->version)) {
C = 16;
}
int W = width / 8;
@@ -1464,10 +1464,10 @@ sd_image_t* txt2img(sd_ctx_t* sd_ctx,
struct ggml_init_params params;
params.mem_size = static_cast<size_t>(10 * 1024 * 1024); // 10 MB
if (sd_ctx->sd->version == VERSION_SD3_2B || sd_ctx->sd->version == VERSION_SD3_5_8B || sd_ctx->sd->version == VERSION_SD3_5_2B) {
if (sd_version_is_sd3(sd_ctx->sd->version)) {
params.mem_size *= 3;
}
if (sd_ctx->sd->version == VERSION_FLUX_DEV || sd_ctx->sd->version == VERSION_FLUX_SCHNELL || sd_ctx->sd->version == VERSION_FLUX_LITE) {
if (sd_version_is_flux(sd_ctx->sd->version)) {
params.mem_size *= 4;
}
if (sd_ctx->sd->stacked_id) {
@@ -1490,17 +1490,17 @@ sd_image_t* txt2img(sd_ctx_t* sd_ctx,
std::vector<float> sigmas = sd_ctx->sd->denoiser->get_sigmas(sample_steps);
int C = 4;
if (sd_ctx->sd->version == VERSION_SD3_2B || sd_ctx->sd->version == VERSION_SD3_5_8B || sd_ctx->sd->version == VERSION_SD3_5_2B) {
if (sd_version_is_sd3(sd_ctx->sd->version)) {
C = 16;
} else if (sd_ctx->sd->version == VERSION_FLUX_DEV || sd_ctx->sd->version == VERSION_FLUX_SCHNELL || sd_ctx->sd->version == VERSION_FLUX_LITE) {
} else if (sd_version_is_flux(sd_ctx->sd->version)) {
C = 16;
}
int W = width / 8;
int H = height / 8;
ggml_tensor* init_latent = ggml_new_tensor_4d(work_ctx, GGML_TYPE_F32, W, H, C, 1);
if (sd_ctx->sd->version == VERSION_SD3_2B || sd_ctx->sd->version == VERSION_SD3_5_8B || sd_ctx->sd->version == VERSION_SD3_5_2B) {
if (sd_version_is_sd3(sd_ctx->sd->version)) {
ggml_set_f32(init_latent, 0.0609f);
} else if (sd_ctx->sd->version == VERSION_FLUX_DEV || sd_ctx->sd->version == VERSION_FLUX_SCHNELL || sd_ctx->sd->version == VERSION_FLUX_LITE) {
} else if (sd_version_is_flux(sd_ctx->sd->version)) {
ggml_set_f32(init_latent, 0.1159f);
} else {
ggml_set_f32(init_latent, 0.f);
@@ -1567,10 +1567,10 @@ sd_image_t* img2img(sd_ctx_t* sd_ctx,
struct ggml_init_params params;
params.mem_size = static_cast<size_t>(10 * 1024 * 1024); // 10 MB
if (sd_ctx->sd->version == VERSION_SD3_2B || sd_ctx->sd->version == VERSION_SD3_5_8B || sd_ctx->sd->version == VERSION_SD3_5_2B) {
if (sd_version_is_sd3(sd_ctx->sd->version)) {
params.mem_size *= 2;
}
if (sd_ctx->sd->version == VERSION_FLUX_DEV || sd_ctx->sd->version == VERSION_FLUX_SCHNELL || sd_ctx->sd->version == VERSION_FLUX_LITE) {
if (sd_version_is_flux(sd_ctx->sd->version)) {
params.mem_size *= 3;
}
if (sd_ctx->sd->stacked_id) {