fix: support more SDXL LoRA names (#216)

* apply pmid lora only once for multiple txt2img calls

* add better support for SDXL LoRA

* fix for some sdxl lora, like lcm-lora-xl

---------

Co-authored-by: bssrdf <[email protected]>
Co-authored-by: leejet <[email protected]>

Author: 3 people

examples/cli/main.cpp

@@ -686,27 +686,26 @@ int main(int argc, const char* argv[]) {
         // Resize input image ...
         if (params.height % 64 != 0 || params.width % 64 != 0) {
             int resized_height = params.height + (64 - params.height % 64);
-            int resized_width = params.width + (64 - params.width % 64);
+            int resized_width  = params.width + (64 - params.width % 64);
 
-            uint8_t *resized_image_buffer = (uint8_t *)malloc(resized_height * resized_width * 3);
+            uint8_t* resized_image_buffer = (uint8_t*)malloc(resized_height * resized_width * 3);
             if (resized_image_buffer == NULL) {
                 fprintf(stderr, "error: allocate memory for resize input image\n");
                 free(input_image_buffer);
                 return 1;
             }
-            stbir_resize(input_image_buffer, params.width, params.height, 0, 
-                resized_image_buffer, resized_width, resized_height, 0, STBIR_TYPE_UINT8, 
-                3 /*RGB channel*/, STBIR_ALPHA_CHANNEL_NONE, 0,
-                STBIR_EDGE_CLAMP, STBIR_EDGE_CLAMP,
-                STBIR_FILTER_BOX, STBIR_FILTER_BOX,
-                STBIR_COLORSPACE_SRGB, nullptr
-            );
+            stbir_resize(input_image_buffer, params.width, params.height, 0,
+                         resized_image_buffer, resized_width, resized_height, 0, STBIR_TYPE_UINT8,
+                         3 /*RGB channel*/, STBIR_ALPHA_CHANNEL_NONE, 0,
+                         STBIR_EDGE_CLAMP, STBIR_EDGE_CLAMP,
+                         STBIR_FILTER_BOX, STBIR_FILTER_BOX,
+                         STBIR_COLORSPACE_SRGB, nullptr);
 
             // Save resized result
             free(input_image_buffer);
             input_image_buffer = resized_image_buffer;
-            params.height = resized_height;
-            params.width = resized_width;
+            params.height      = resized_height;
+            params.width       = resized_width;
         }
     }
 

lora.hpp

@@ -11,7 +11,7 @@ struct LoraModel : public GGMLModule {
     std::string file_path;
     ModelLoader model_loader;
     bool load_failed = false;
-    bool applied = false;
+    bool applied     = false;
 
     LoraModel(ggml_backend_t backend,
               ggml_type wtype,
@@ -91,10 +91,15 @@ struct LoraModel : public GGMLModule {
             k_tensor = k_tensor.substr(0, k_pos);
             replace_all_chars(k_tensor, '.', '_');
             // LOG_DEBUG("k_tensor %s", k_tensor.c_str());
-            if (k_tensor == "model_diffusion_model_output_blocks_2_2_conv") {  // fix for SDXL
-                k_tensor = "model_diffusion_model_output_blocks_2_1_conv";
+            std::string lora_up_name = "lora." + k_tensor + ".lora_up.weight";
+            if (lora_tensors.find(lora_up_name) == lora_tensors.end()) {
+                if (k_tensor == "model_diffusion_model_output_blocks_2_2_conv") {
+                    // fix for some sdxl lora, like lcm-lora-xl
+                    k_tensor     = "model_diffusion_model_output_blocks_2_1_conv";
+                    lora_up_name = "lora." + k_tensor + ".lora_up.weight";
+                }
             }
-            std::string lora_up_name   = "lora." + k_tensor + ".lora_up.weight";
+
             std::string lora_down_name = "lora." + k_tensor + ".lora_down.weight";
             std::string alpha_name     = "lora." + k_tensor + ".alpha";
             std::string scale_name     = "lora." + k_tensor + ".scale";

model.cpp

@@ -211,6 +211,8 @@ std::string convert_sdxl_lora_name(std::string tensor_name) {
         {"unet", "model_diffusion_model"},
         {"te2", "cond_stage_model_1_transformer"},
         {"te1", "cond_stage_model_transformer"},
+        {"text_encoder_2", "cond_stage_model_1_transformer"},
+        {"text_encoder", "cond_stage_model_transformer"},
     };
     for (auto& pair_i : sdxl_lora_name_lookup) {
         if (tensor_name.compare(0, pair_i.first.length(), pair_i.first) == 0) {
@@ -446,18 +448,25 @@ std::string convert_tensor_name(const std::string& name) {
         } else {
             new_name = name;
         }
-    } else if (contains(name, "lora_up") || contains(name, "lora_down") || contains(name, "lora.up") || contains(name, "lora.down")) {
+    } else if (contains(name, "lora_up") || contains(name, "lora_down") ||
+               contains(name, "lora.up") || contains(name, "lora.down") ||
+               contains(name, "lora_linear")) {
         size_t pos = new_name.find(".processor");
         if (pos != std::string::npos) {
             new_name.replace(pos, strlen(".processor"), "");
         }
-        pos = new_name.find_last_of('_');
+        pos = new_name.rfind("lora");
         if (pos != std::string::npos) {
-            std::string name_without_network_parts = new_name.substr(0, pos);
-            std::string network_part               = new_name.substr(pos + 1);
+            std::string name_without_network_parts = new_name.substr(0, pos - 1);
+            std::string network_part               = new_name.substr(pos);
             // LOG_DEBUG("%s %s", name_without_network_parts.c_str(), network_part.c_str());
             std::string new_key = convert_diffusers_name_to_compvis(name_without_network_parts, '.');
+            new_key             = convert_sdxl_lora_name(new_key);
             replace_all_chars(new_key, '.', '_');
+            size_t npos = network_part.rfind("_linear_layer");
+            if (npos != std::string::npos) {
+                network_part.replace(npos, strlen("_linear_layer"), "");
+            }
             if (starts_with(network_part, "lora.")) {
                 network_part = "lora_" + network_part.substr(5);
             }

stable-diffusion.cpp

@@ -1610,7 +1610,7 @@ sd_image_t* txt2img(sd_ctx_t* sd_ctx,
     if (sd_ctx->sd->stacked_id && !sd_ctx->sd->pmid_lora->applied) {
         t0 = ggml_time_ms();
         sd_ctx->sd->pmid_lora->apply(sd_ctx->sd->tensors, sd_ctx->sd->n_threads);
-        t1 = ggml_time_ms();
+        t1                             = ggml_time_ms();
         sd_ctx->sd->pmid_lora->applied = true;
         LOG_INFO("pmid_lora apply completed, taking %.2fs", (t1 - t0) * 1.0f / 1000);
         if (sd_ctx->sd->free_params_immediately) {