ggml

@@ -1 +1 @@
-Subproject commit 2d3876d554551d35c06dccc5852be50d5fd2a275
+Subproject commit c538174d261d8172480f87efcfec8e69aac13ebb

ggml_extend.hpp

@@ -954,16 +954,7 @@ __STATIC_INLINE__ struct ggml_tensor* ggml_nn_linear(struct ggml_context* ctx,
     if (scale != 1.f) {
         x = ggml_scale(ctx, x, scale);
     }
-    if (x->ne[2] * x->ne[3] > 1024) {
-        // workaround: avoid ggml cuda error
-        int64_t ne2 = x->ne[2];
-        int64_t ne3 = x->ne[3];
-        x           = ggml_reshape_2d(ctx, x, x->ne[0], x->ne[1] * x->ne[2] * x->ne[3]);
-        x           = ggml_mul_mat(ctx, w, x);
-        x           = ggml_reshape_4d(ctx, x, x->ne[0], x->ne[1] / ne2 / ne3, ne2, ne3);
-    } else {
-        x = ggml_mul_mat(ctx, w, x);
-    }
+    x = ggml_mul_mat(ctx, w, x);
     if (force_prec_f32) {
         ggml_mul_mat_set_prec(x, GGML_PREC_F32);
     }

stable-diffusion.cpp

@@ -765,14 +765,11 @@ public:
                 denoiser = std::make_shared<DiscreteFlowDenoiser>(shift);
             } else if (sd_version_is_flux(version)) {
                 LOG_INFO("running in Flux FLOW mode");
-                float shift = sd_ctx_params->flow_shift;
-                if (shift == INFINITY) {
-                    shift = 1.0f;  // TODO: validate
-                    for (auto pair : model_loader.tensor_storages_types) {
-                        if (pair.first.find("model.diffusion_model.guidance_in.in_layer.weight") != std::string::npos) {
-                            shift = 1.15f;
-                            break;
-                        }
+                float shift = 1.0f;  // TODO: validate
+                for (auto pair : model_loader.tensor_storages_types) {
+                    if (pair.first.find("model.diffusion_model.guidance_in.in_layer.weight") != std::string::npos) {
+                        shift = 1.15f;
+                        break;
                     }
                 }
                 denoiser = std::make_shared<FluxFlowDenoiser>(shift);