patch size consistent with Flux1

Author: stduhpf

flux.hpp

@@ -1250,7 +1250,6 @@ namespace Flux {
             } else if (sd_version_is_longcat(version)) {
                 flux_params.context_in_dim = 3584;
                 flux_params.vec_in_dim     = 0;
-                flux_params.patch_size     = 1;
             }
             for (auto pair : tensor_storage_map) {
                 std::string tensor_name = pair.first;

stable-diffusion.cpp

@@ -1346,12 +1346,6 @@ class StableDiffusionGGML {
                     latent_rgb_bias = flux2_latent_rgb_bias;
                     patch_sz        = 2;
                 }
-            } else if (dim == 64) {
-                if (sd_version_is_flux(version) || sd_version_is_z_image(version) || sd_version_is_longcat(version)) {
-                    latent_rgb_proj = flux_latent_rgb_proj;
-                    latent_rgb_bias = flux_latent_rgb_bias;
-                    patch_sz        = 2;
-                }
             } else if (dim == 48) {
                 if (sd_version_is_wan(version)) {
                     latent_rgb_proj = wan_22_latent_rgb_proj;
@@ -1921,7 +1915,7 @@ class StableDiffusionGGML {
         int vae_scale_factor = 8;
         if (version == VERSION_WAN2_2_TI2V) {
             vae_scale_factor = 16;
-        } else if (sd_version_is_flux2(version) || sd_version_is_longcat(version)) {
+        } else if (sd_version_is_flux2(version)) {
             vae_scale_factor = 16;
         } else if (version == VERSION_CHROMA_RADIANCE) {
             vae_scale_factor = 1;
@@ -1938,8 +1932,6 @@ class StableDiffusionGGML {
                 latent_channel = 3;
             } else if (sd_version_is_flux2(version)) {
                 latent_channel = 128;
-            } else if (sd_version_is_longcat(version)) {
-                latent_channel = 64;
             } else {
                 latent_channel = 16;
             }
@@ -2240,7 +2232,6 @@ class StableDiffusionGGML {
             sd_version_is_qwen_image(version) ||
             sd_version_is_wan(version) ||
             sd_version_is_flux2(version) ||
-            sd_version_is_longcat(version) ||
             version == VERSION_CHROMA_RADIANCE) {
             latent = vae_output;
         } else if (version == VERSION_SD1_PIX2PIX) {

vae.hpp

@@ -553,7 +553,7 @@ class AutoencodingEngine : public GGMLBlock {
 
     struct ggml_tensor* decode(GGMLRunnerContext* ctx, struct ggml_tensor* z) {
         // z: [N, z_channels, h, w]
-        if (sd_version_is_flux2(version) || sd_version_is_longcat(version)) {
+        if (sd_version_is_flux2(version)) {
             // [N, C*p*p, h, w] -> [N, C, h*p, w*p]
             int64_t p = 2;
 
@@ -592,7 +592,7 @@ class AutoencodingEngine : public GGMLBlock {
             auto quant_conv = std::dynamic_pointer_cast<Conv2d>(blocks["quant_conv"]);
             z               = quant_conv->forward(ctx, z);  // [N, 2*embed_dim, h/8, w/8]
         }
-        if (sd_version_is_flux2(version) || sd_version_is_longcat(version)) {
+        if (sd_version_is_flux2(version)) {
             z = ggml_ext_chunk(ctx->ggml_ctx, z, 2, 2)[0];
 
             // [N, C, H, W] -> [N, C*p*p, H/p, W/p]