DarthAffe/stable-diffusion.cpp
1
0
Fork 0
mirror of https://github.com/leejet/stable-diffusion.cpp.git synced 2026-08-10 14:16:50 +00:00
Code Issues Packages Projects Releases Wiki Activity

fix: avoid f16 overflow in Z-Image quantized matmuls on ROCm (#1771)

Browse Source
This commit is contained in:
Piotr Wilkin (ilintar) 2026-07-10 17:22:15 +02:00 committed by GitHub
parent 12b6fbff28
commit 9beb6aca69
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 3 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
src/model/diffusion/z_image.hpp
View File

@ -150,6 +150,8 @@ namespace ZImage {
if (sd_backend_is(ctx->backend, "ROCm")) { if (sd_backend_is(ctx->backend, "ROCm")) {
out_proj->set_scale(1.f / 16.f); out_proj->set_scale(1.f / 16.f);
out_proj->set_force_prec_f32(true);
qkv_proj->set_force_prec_f32(true);
} }
auto qkv = qkv_proj->forward(ctx, x); // [N, n_token, (num_heads + num_kv_heads*2)*head_dim] auto qkv = qkv_proj->forward(ctx, x); // [N, n_token, (num_heads + num_kv_heads*2)*head_dim]
@ -227,7 +229,7 @@ namespace ZImage {
auto w2 = std::dynamic_pointer_cast<Linear>(blocks["w2"]); auto w2 = std::dynamic_pointer_cast<Linear>(blocks["w2"]);
auto w3 = std::dynamic_pointer_cast<Linear>(blocks["w3"]); auto w3 = std::dynamic_pointer_cast<Linear>(blocks["w3"]);
if (sd_backend_is(ctx->backend, "Vulkan")) { if (sd_backend_is(ctx->backend, "Vulkan") || sd_backend_is(ctx->backend, "ROCm")) {
w2->set_force_prec_f32(true); w2->set_force_prec_f32(true);
} }