From 26f591db6d9fedb111ccc816a511310f855284c0 Mon Sep 17 00:00:00 2001
From: bssrdf <bssrdf@gmail.com>
Date: Thu, 15 Feb 2024 21:56:14 -0500
Subject: [PATCH] disable pmid lora apply for now; 1 input image seems working;
 > 1 not working

---
 pmid.hpp             | 3 ++-
 stable-diffusion.cpp | 4 ++--
 2 files changed, 4 insertions(+), 3 deletions(-)

diff --git a/pmid.hpp b/pmid.hpp
index a4c2ea98..8a056f15 100644
--- a/pmid.hpp
+++ b/pmid.hpp
@@ -234,6 +234,7 @@ struct FuseModule{
         // print_ggml_tensor(prompt_embeds, true, "prompt_embeds_perm");
         // class_tokens_mask = ggml_repeat(ctx, class_tokens_mask, prompt_embeds_perm);
         class_tokens_mask = ggml_cont(ctx, ggml_transpose(ctx, class_tokens_mask));
+        class_tokens_mask = ggml_repeat(ctx, class_tokens_mask, prompt_embeds);
         // print_ggml_tensor(class_tokens_mask, true, "class_tokens_mask_repeat");
         // class_tokens_mask =  ggml_cont(ctx, ggml_transpose(ctx, class_tokens_mask));
         // print_ggml_tensor(class_tokens_mask, true, "class_tokens_mask_transpose");
@@ -334,7 +335,7 @@ struct PhotoMakerIDEncoder : public GGMLModule {
                                                              class_embedding_temp,
                                                              positions
                                                              ); // [batch_size, seq_length, hidden_size]
-        print_ggml_tensor(shared_id_embeds, true, "shared_id_embeds");
+        // print_ggml_tensor(shared_id_embeds, true, "shared_id_embeds");
 
         // if(class_tokens_mask->backend == GGML_BACKEND_GPU){            
         //     int *ctm = (int *)malloc(class_tokens_mask->ne[0]);
diff --git a/stable-diffusion.cpp b/stable-diffusion.cpp
index 4da9fd85..c999d0b0 100644
--- a/stable-diffusion.cpp
+++ b/stable-diffusion.cpp
@@ -1471,8 +1471,8 @@ sd_image_t* txt2img(sd_ctx_t* sd_ctx,
     LOG_INFO("apply_loras completed, taking %.2fs", (t1 - t0) * 1.0f / 1000);
 
     if(sd_ctx->sd->stacked_id){
-        sd_ctx->sd->pmid_lora.apply(sd_ctx->sd->tensors, sd_ctx->sd->n_threads);
-        sd_ctx->sd->pmid_lora.free_compute_buffer();
+        // sd_ctx->sd->pmid_lora.apply(sd_ctx->sd->tensors, sd_ctx->sd->n_threads);
+        // sd_ctx->sd->pmid_lora.free_compute_buffer();
         if (sd_ctx->sd->free_params_immediately) {
             sd_ctx->sd->pmid_lora.free_params_buffer();
         }