Skip to content

Avoid generating black images when running T5 on the GPU #881

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 1 commit into from
Oct 12, 2025
Merged

Avoid generating black images when running T5 on the GPU #881

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
12 changes: 1 addition & 11 deletions stable-diffusion.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -338,17 +338,7 @@ class StableDiffusionGGML {
bool clip_on_cpu = sd_ctx_params->keep_clip_on_cpu;

{
clip_backend = backend;
bool use_t5xxl = false;
if (sd_version_is_dit(version) && !sd_version_is_qwen_image(version)) {
use_t5xxl = true;
}
if (!clip_on_cpu && !ggml_backend_is_cpu(backend) && use_t5xxl) {
LOG_WARN(
"!!!It appears that you are using the T5 model. Some backends may encounter issues with it."
"If you notice that the generated images are completely black,"
"try running the T5 model on the CPU using the --clip-on-cpu parameter.");
}
clip_backend = backend;
if (clip_on_cpu && !ggml_backend_is_cpu(backend)) {
LOG_INFO("CLIP: Using CPU backend");
clip_backend = ggml_backend_cpu_init();
Expand Down
4 changes: 3 additions & 1 deletion t5.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -504,7 +504,9 @@ struct T5DenseGatedActDense : public UnaryBlock {
T5DenseGatedActDense(int64_t model_dim, int64_t ff_dim) {
blocks["wi_0"] = std::shared_ptr<GGMLBlock>(new Linear(model_dim, ff_dim, false));
blocks["wi_1"] = std::shared_ptr<GGMLBlock>(new Linear(model_dim, ff_dim, false));
blocks["wo"] = std::shared_ptr<GGMLBlock>(new Linear(ff_dim, model_dim, false));
float scale = 1.f / 32.f;
// The purpose of the scale here is to prevent NaN issues on some backends(CUDA, ...).
blocks["wo"] = std::shared_ptr<GGMLBlock>(new Linear(ff_dim, model_dim, false, false, false, scale));
}

struct ggml_tensor* forward(struct ggml_context* ctx, struct ggml_tensor* x) {
Expand Down
Loading