feat: support JPEG compression (#583)

Signed-off-by: Xiaodong Ye <xiaodong.ye@mthreads.com>

.github/workflows/build.yml

@@ -155,15 +155,15 @@ jobs:
       matrix:
         include:
           - build: "noavx"
-            defines: "-DGGML_AVX=OFF -DGGML_AVX2=OFF -DGGML_FMA=OFF -DSD_BUILD_SHARED_LIBS=ON"
+            defines: "-DGGML_NATIVE=OFF -DGGML_AVX=OFF -DGGML_AVX2=OFF -DGGML_FMA=OFF -DSD_BUILD_SHARED_LIBS=ON"
           - build: "avx2"
-            defines: "-DGGML_AVX2=ON -DSD_BUILD_SHARED_LIBS=ON"
+            defines: "-DGGML_NATIVE=OFF -DGGML_AVX2=ON -DSD_BUILD_SHARED_LIBS=ON"
           - build: "avx"
-            defines: "-DGGML_AVX2=OFF -DSD_BUILD_SHARED_LIBS=ON"
+            defines: "-DGGML_NATIVE=OFF -DGGML_AVX=ON -DGGML_AVX2=OFF -DSD_BUILD_SHARED_LIBS=ON"
           - build: "avx512"
-            defines: "-DGGML_AVX512=ON -DSD_BUILD_SHARED_LIBS=ON"
+            defines: "-DGGML_NATIVE=OFF -DGGML_AVX512=ON -DGGML_AVX=ON -DGGML_AVX2=ON -DSD_BUILD_SHARED_LIBS=ON"
           - build: "cuda12"
-            defines: "-DSD_CUBLAS=ON -DSD_BUILD_SHARED_LIBS=ON"
+            defines: "-DSD_CUDA=ON -DSD_BUILD_SHARED_LIBS=ON -DCMAKE_CUDA_ARCHITECTURES=90;89;80;75"
           # - build: "rocm5.5"
           #   defines: '-G Ninja -DCMAKE_C_COMPILER=clang -DCMAKE_CXX_COMPILER=clang++ -DSD_HIPBLAS=ON -DCMAKE_BUILD_TYPE=Release -DAMDGPU_TARGETS="gfx1100;gfx1102;gfx1030" -DSD_BUILD_SHARED_LIBS=ON'
           - build: 'vulkan'
@@ -178,9 +178,9 @@ jobs:
       - name: Install cuda-toolkit
         id: cuda-toolkit
         if: ${{ matrix.build == 'cuda12' }}
-        uses: Jimver/cuda-toolkit@v0.2.11
+        uses: Jimver/cuda-toolkit@v0.2.19
         with:
-          cuda: "12.2.0"
+          cuda: "12.6.2"
           method: "network"
           sub-packages: '["nvcc", "cudart", "cublas", "cublas_dev", "thrust", "visual_studio_integration"]'
 

CMakeLists.txt

@@ -24,19 +24,20 @@ endif()
 # general
 #option(SD_BUILD_TESTS                "sd: build tests"    ${SD_STANDALONE})
 option(SD_BUILD_EXAMPLES             "sd: build examples" ${SD_STANDALONE})
-option(SD_CUBLAS                     "sd: cuda backend" OFF)
+option(SD_CUDA                       "sd: cuda backend" OFF)
 option(SD_HIPBLAS                    "sd: rocm backend" OFF)
 option(SD_METAL                      "sd: metal backend" OFF)
 option(SD_VULKAN                     "sd: vulkan backend" OFF)
 option(SD_SYCL                       "sd: sycl backend" OFF)
+option(SD_MUSA                       "sd: musa backend" OFF)
 option(SD_FAST_SOFTMAX               "sd: x1.5 faster softmax, indeterministic (sometimes, same seed don't generate same image), cuda only" OFF)
 option(SD_BUILD_SHARED_LIBS          "sd: build shared libs" OFF)
 #option(SD_BUILD_SERVER               "sd: build server example"                           ON)
 
-if(SD_CUBLAS)
-    message("-- Use CUBLAS as backend stable-diffusion")
+if(SD_CUDA)
+    message("-- Use CUDA as backend stable-diffusion")
     set(GGML_CUDA ON)
-    add_definitions(-DSD_USE_CUBLAS)
+    add_definitions(-DSD_USE_CUDA)
 endif()
 
 if(SD_METAL)
@@ -53,16 +54,25 @@ endif ()
 
 if (SD_HIPBLAS)
     message("-- Use HIPBLAS as backend stable-diffusion")
-    set(GGML_HIPBLAS ON)
-    add_definitions(-DSD_USE_CUBLAS)
+    set(GGML_HIP ON)
+    add_definitions(-DSD_USE_CUDA)
     if(SD_FAST_SOFTMAX)
         set(GGML_CUDA_FAST_SOFTMAX ON)
     endif()
 endif ()
 
+if(SD_MUSA)
+    message("-- Use MUSA as backend stable-diffusion")
+    set(GGML_MUSA ON)
+    add_definitions(-DSD_USE_CUDA)
+    if(SD_FAST_SOFTMAX)
+        set(GGML_CUDA_FAST_SOFTMAX ON)
+    endif()
+endif()
+
 set(SD_LIB stable-diffusion)
 
-file(GLOB SD_LIB_SOURCES 
+file(GLOB SD_LIB_SOURCES
     "*.h"
     "*.cpp"
     "*.hpp"

Dockerfile.musa

@@ -0,0 +1,19 @@
+ARG MUSA_VERSION=rc3.1.0
+
+FROM mthreads/musa:${MUSA_VERSION}-devel-ubuntu22.04 as build
+
+RUN apt-get update && apt-get install -y cmake
+
+WORKDIR /sd.cpp
+
+COPY . .
+
+RUN mkdir build && cd build && \
+    cmake .. -DCMAKE_C_COMPILER=clang -DCMAKE_CXX_COMPILER=clang++ -DSD_MUSA=ON -DCMAKE_BUILD_TYPE=Release && \
+    cmake --build . --config Release
+
+FROM mthreads/musa:${MUSA_VERSION}-runtime-ubuntu22.04 as runtime
+
+COPY --from=build /sd.cpp/build/bin/sd /sd
+
+ENTRYPOINT [ "/sd" ]

README.md

@@ -113,12 +113,12 @@ cmake .. -DGGML_OPENBLAS=ON
 cmake --build . --config Release
 ```
 
-##### Using CUBLAS
+##### Using CUDA
 
 This provides BLAS acceleration using the CUDA cores of your Nvidia GPU. Make sure to have the CUDA toolkit installed. You can download it from your Linux distro's package manager (e.g. `apt install nvidia-cuda-toolkit`) or from here: [CUDA Toolkit](https://developer.nvidia.com/cuda-downloads). Recommended to have at least 4 GB of VRAM.
 
 ```
-cmake .. -DSD_CUBLAS=ON
+cmake .. -DSD_CUDA=ON
 cmake --build . --config Release
 ```
 
@@ -132,6 +132,14 @@ cmake .. -G "Ninja" -DCMAKE_C_COMPILER=clang -DCMAKE_CXX_COMPILER=clang++ -DSD_H
 cmake --build . --config Release
 ```
 
+##### Using MUSA
+
+This provides BLAS acceleration using the MUSA cores of your Moore Threads GPU. Make sure to have the MUSA toolkit installed.
+
+```bash
+cmake .. -DCMAKE_C_COMPILER=/usr/local/musa/bin/clang -DCMAKE_CXX_COMPILER=/usr/local/musa/bin/clang++ -DSD_MUSA=ON -DCMAKE_BUILD_TYPE=Release
+cmake --build . --config Release
+```
 
 ##### Using Metal
 
@@ -232,6 +240,10 @@ arguments:
   -p, --prompt [PROMPT]              the prompt to render
   -n, --negative-prompt PROMPT       the negative prompt (default: "")
   --cfg-scale SCALE                  unconditional guidance scale: (default: 7.0)
+  --skip-layers LAYERS               Layers to skip for SLG steps: (default: [7,8,9])
+  --skip-layer-start START           SLG enabling point: (default: 0.01)
+  --skip-layer-end END               SLG disabling point: (default: 0.2)
+									 SLG will be enabled at step int([STEPS]*[START]) and disabled at int([STEPS]*[END])
   --strength STRENGTH                strength for noising/unnoising (default: 0.75)
   --style-ratio STYLE-RATIO          strength for keeping input identity (default: 20%)
   --control-strength STRENGTH        strength to apply Control Net (default: 0.9)
@@ -314,6 +326,7 @@ These projects use `stable-diffusion.cpp` as a backend for their image generatio
 
 - [Jellybox](https://jellybox.com)
 - [Stable Diffusion GUI](https://github.com/fszontagh/sd.cpp.gui.wx)
+- [Stable Diffusion CLI-GUI] (https://github.com/piallai/stable-diffusion.cpp)
 
 ## Contributors
 

examples/cli/main.cpp

@@ -209,9 +209,9 @@ void print_usage(int argc, const char* argv[]) {
     printf("  --cfg-scale SCALE                  unconditional guidance scale: (default: 7.0)\n");
     printf("  --slg-scale SCALE                  skip layer guidance (SLG) scale, only for DiT models: (default: 0)\n");
     printf("                                     0 means disabled, a value of 2.5 is nice for sd3.5 medium\n");
-    printf("  --skip_layers LAYERS               Layers to skip for SLG steps: (default: [7,8,9])\n");
-    printf("  --skip_layer_start START           SLG enabling point: (default: 0.01)\n");
-    printf("  --skip_layer_end END               SLG disabling point: (default: 0.2)\n");
+    printf("  --skip-layers LAYERS               Layers to skip for SLG steps: (default: [7,8,9])\n");
+    printf("  --skip-layer-start START           SLG enabling point: (default: 0.01)\n");
+    printf("  --skip-layer-end END               SLG disabling point: (default: 0.2)\n");
     printf("                                     SLG will be enabled at step int([STEPS]*[START]) and disabled at int([STEPS]*[END])\n");
     printf("  --strength STRENGTH                strength for noising/unnoising (default: 0.75)\n");
     printf("  --style-ratio STYLE-RATIO          strength for keeping input identity (default: 20%%)\n");
@@ -1057,16 +1057,41 @@ int main(int argc, const char* argv[]) {
         }
     }
 
-    size_t last            = params.output_path.find_last_of(".");
-    std::string dummy_name = last != std::string::npos ? params.output_path.substr(0, last) : params.output_path;
+    std::string dummy_name, ext, lc_ext;
+    bool is_jpg;
+    size_t last = params.output_path.find_last_of(".");
+    size_t last_path = std::min(params.output_path.find_last_of("/"),
+                                params.output_path.find_last_of("\\"));
+    if (last != std::string::npos // filename has extension
+    && (last_path == std::string::npos || last > last_path)) {
+        dummy_name = params.output_path.substr(0, last);
+        ext = lc_ext = params.output_path.substr(last);
+        std::transform(ext.begin(), ext.end(), lc_ext.begin(), ::tolower);
+        is_jpg = lc_ext == ".jpg" || lc_ext == ".jpeg" || lc_ext == ".jpe";
+    } else {
+        dummy_name = params.output_path;
+        ext = lc_ext = "";
+        is_jpg = false;
+    }
+    // appending ".png" to absent or unknown extension
+    if (!is_jpg && lc_ext != ".png") {
+        dummy_name += ext;
+        ext = ".png";
+    }
     for (int i = 0; i < params.batch_count; i++) {
         if (results[i].data == NULL) {
             continue;
         }
-        std::string final_image_path = i > 0 ? dummy_name + "_" + std::to_string(i + 1) + ".png" : dummy_name + ".png";
-        stbi_write_png(final_image_path.c_str(), results[i].width, results[i].height, results[i].channel,
-                       results[i].data, 0, get_image_params(params, params.seed + i).c_str());
-        printf("save result image to '%s'\n", final_image_path.c_str());
+        std::string final_image_path = i > 0 ? dummy_name + "_" + std::to_string(i + 1) + ext : dummy_name + ext;
+        if(is_jpg) {
+            stbi_write_jpg(final_image_path.c_str(), results[i].width, results[i].height, results[i].channel,
+                           results[i].data, 90, get_image_params(params, params.seed + i).c_str());
+            printf("save result JPEG image to '%s'\n", final_image_path.c_str());
+        } else {
+            stbi_write_png(final_image_path.c_str(), results[i].width, results[i].height, results[i].channel,
+                           results[i].data, 0, get_image_params(params, params.seed + i).c_str());
+            printf("save result PNG image to '%s'\n", final_image_path.c_str());
+        }
         free(results[i].data);
         results[i].data = NULL;
     }

ggml_extend.hpp

@@ -27,7 +27,7 @@
 
 #include "model.h"
 
-#ifdef SD_USE_CUBLAS
+#ifdef SD_USE_CUDA
 #include "ggml-cuda.h"
 #endif
 
@@ -708,7 +708,7 @@ __STATIC_INLINE__ struct ggml_tensor* ggml_nn_attention(struct ggml_context* ctx
                                                         struct ggml_tensor* k,
                                                         struct ggml_tensor* v,
                                                         bool mask = false) {
-#if defined(SD_USE_FLASH_ATTENTION) && !defined(SD_USE_CUBLAS) && !defined(SD_USE_METAL) && !defined(SD_USE_VULKAN) && !defined(SD_USE_SYCL)
+#if defined(SD_USE_FLASH_ATTENTION) && !defined(SD_USE_CUDA) && !defined(SD_USE_METAL) && !defined(SD_USE_VULKAN) && !defined(SD_USE_SYCL)
     struct ggml_tensor* kqv = ggml_flash_attn(ctx, q, k, v, false);  // [N * n_head, n_token, d_head]
 #else
     float d_head           = (float)q->ne[0];
@@ -864,7 +864,7 @@ __STATIC_INLINE__ struct ggml_tensor* ggml_nn_group_norm(struct ggml_context* ct
 }
 
 __STATIC_INLINE__ void ggml_backend_tensor_get_and_sync(ggml_backend_t backend, const struct ggml_tensor* tensor, void* data, size_t offset, size_t size) {
-#if defined(SD_USE_CUBLAS) || defined(SD_USE_SYCL)
+#if defined(SD_USE_CUDA) || defined(SD_USE_SYCL)
     if (!ggml_backend_is_cpu(backend)) {
         ggml_backend_tensor_get_async(backend, tensor, data, offset, size);
         ggml_backend_synchronize(backend);

lora.hpp

@@ -615,9 +615,12 @@ struct LoraModel : public GGMLRunner {
                 scale_value *= multiplier;
 
                 // flat lora tensors to multiply it
-                int64_t lora_up_rows   = lora_up->ne[ggml_n_dims(lora_up) - 1];
-                lora_up                = ggml_reshape_2d(compute_ctx, lora_up, ggml_nelements(lora_up) / lora_up_rows, lora_up_rows);
-                int64_t lora_down_rows = lora_down->ne[ggml_n_dims(lora_down) - 1];
+                int64_t lora_up_rows  = lora_up->ne[ggml_n_dims(lora_up) - 1];
+                lora_up               = ggml_reshape_2d(compute_ctx, lora_up, ggml_nelements(lora_up) / lora_up_rows, lora_up_rows);
+                auto lora_down_n_dims = ggml_n_dims(lora_down);
+                // assume n_dims should always be a multiple of 2 (otherwise rank 1 doesn't work)
+                lora_down_n_dims       = (lora_down_n_dims + lora_down_n_dims % 2);
+                int64_t lora_down_rows = lora_down->ne[lora_down_n_dims - 1];
                 lora_down              = ggml_reshape_2d(compute_ctx, lora_down, ggml_nelements(lora_down) / lora_down_rows, lora_down_rows);
 
                 // ggml_mul_mat requires tensor b transposed

model.cpp

@@ -1748,9 +1748,11 @@ bool ModelLoader::load_tensors(on_new_tensor_cb_t on_new_tensor_cb, ggml_backend
             }
             return true;
         };
-
+        int tensor_count = 0;
+        int64_t t1       = ggml_time_ms();
         for (auto& tensor_storage : processed_tensor_storages) {
             if (tensor_storage.file_index != file_index) {
+                ++tensor_count;
                 continue;
             }
             ggml_tensor* dst_tensor = NULL;
@@ -1762,6 +1764,7 @@ bool ModelLoader::load_tensors(on_new_tensor_cb_t on_new_tensor_cb, ggml_backend
             }
 
             if (dst_tensor == NULL) {
+                ++tensor_count;
                 continue;
             }
 
@@ -1828,6 +1831,9 @@ bool ModelLoader::load_tensors(on_new_tensor_cb_t on_new_tensor_cb, ggml_backend
                     ggml_backend_tensor_set(dst_tensor, convert_buffer.data(), 0, ggml_nbytes(dst_tensor));
                 }
             }
+            int64_t t2 = ggml_time_ms();
+            pretty_progress(++tensor_count, processed_tensor_storages.size(), (t2 - t1) / 1000.0f);
+            t1 = t2;
         }
 
         if (zip != NULL) {

stable-diffusion.cpp

@@ -159,7 +159,7 @@ public:
                         bool vae_on_cpu,
                         bool diffusion_flash_attn) {
         use_tiny_autoencoder = taesd_path.size() > 0;
-#ifdef SD_USE_CUBLAS
+#ifdef SD_USE_CUDA
         LOG_DEBUG("Using CUDA backend");
         backend = ggml_backend_cuda_init(0);
 #endif
@@ -360,7 +360,7 @@ public:
                 first_stage_model->alloc_params_buffer();
                 first_stage_model->get_param_tensors(tensors, "first_stage_model");
             } else {
-                tae_first_stage = std::make_shared<TinyAutoEncoder>(backend, model_loader.tensor_storages_types, "decoder.layers", vae_decode_only);
+                tae_first_stage = std::make_shared<TinyAutoEncoder>(backend, model_loader.tensor_storages_types, "decoder.layers", vae_decode_only, version);
             }
             // first_stage_model->get_param_tensors(tensors, "first_stage_model.");
 
@@ -525,6 +525,10 @@ public:
             if (is_using_v_parameterization_for_sd2(ctx, sd_version_is_inpaint(version))) {
                 is_using_v_parameterization = true;
             }
+        } else if (sd_version_is_sdxl(version)) {
+            if (model_loader.tensor_storages_types.find("v_pred") != model_loader.tensor_storages_types.end()) {
+                is_using_v_parameterization = true;
+            }
         } else if (version == VERSION_SVD) {
             // TODO: V_PREDICTION_EDM
             is_using_v_parameterization = true;
@@ -606,7 +610,9 @@ public:
         ggml_set_f32(timesteps, 999);
 
         struct ggml_tensor* concat = is_inpaint ? ggml_new_tensor_4d(work_ctx, GGML_TYPE_F32, 8, 8, 5, 1) : NULL;
-        ggml_set_f32(concat, 0);
+        if (concat != NULL) {
+            ggml_set_f32(concat, 0);
+        }
 
         int64_t t0              = ggml_time_ms();
         struct ggml_tensor* out = ggml_dup_tensor(work_ctx, x_t);

tae.hpp

@@ -62,7 +62,8 @@ class TinyEncoder : public UnaryBlock {
     int num_blocks  = 3;
 
 public:
-    TinyEncoder() {
+    TinyEncoder(int z_channels = 4)
+        : z_channels(z_channels) {
         int index                       = 0;
         blocks[std::to_string(index++)] = std::shared_ptr<GGMLBlock>(new Conv2d(in_channels, channels, {3, 3}, {1, 1}, {1, 1}));
         blocks[std::to_string(index++)] = std::shared_ptr<GGMLBlock>(new TAEBlock(channels, channels));
@@ -106,7 +107,10 @@ class TinyDecoder : public UnaryBlock {
     int num_blocks   = 3;
 
 public:
-    TinyDecoder(int index = 0) {
+    TinyDecoder(int z_channels = 4)
+        : z_channels(z_channels) {
+        int index = 0;
+
         blocks[std::to_string(index++)] = std::shared_ptr<GGMLBlock>(new Conv2d(z_channels, channels, {3, 3}, {1, 1}, {1, 1}));
         index++;  // nn.ReLU()
 
@@ -163,12 +167,16 @@ protected:
     bool decode_only;
 
 public:
-    TAESD(bool decode_only = true)
+    TAESD(bool decode_only = true, SDVersion version = VERSION_SD1)
         : decode_only(decode_only) {
-        blocks["decoder.layers"] = std::shared_ptr<GGMLBlock>(new TinyDecoder());
+        int z_channels = 4;
+        if (sd_version_is_dit(version)) {
+            z_channels = 16;
+        }
+        blocks["decoder.layers"] = std::shared_ptr<GGMLBlock>(new TinyDecoder(z_channels));
 
         if (!decode_only) {
-            blocks["encoder.layers"] = std::shared_ptr<GGMLBlock>(new TinyEncoder());
+            blocks["encoder.layers"] = std::shared_ptr<GGMLBlock>(new TinyEncoder(z_channels));
         }
     }
 
@@ -190,9 +198,10 @@ struct TinyAutoEncoder : public GGMLRunner {
     TinyAutoEncoder(ggml_backend_t backend,
                     std::map<std::string, enum ggml_type>& tensor_types,
                     const std::string prefix,
-                    bool decoder_only = true)
+                    bool decoder_only = true,
+                    SDVersion version = VERSION_SD1)
         : decode_only(decoder_only),
-          taesd(decode_only),
+          taesd(decode_only, version),
           GGMLRunner(backend) {
         taesd.init(params_ctx, tensor_types, prefix);
     }

thirdparty/stb_image_write.h

@@ -1412,7 +1412,7 @@ static int stbiw__jpg_processDU(stbi__write_context *s, int *bitBuf, int *bitCnt
    return DU[0];
 }
 
-static int stbi_write_jpg_core(stbi__write_context *s, int width, int height, int comp, const void* data, int quality) {
+static int stbi_write_jpg_core(stbi__write_context *s, int width, int height, int comp, const void* data, int quality, const char* parameters) {
    // Constants that don't pollute global namespace
    static const unsigned char std_dc_luminance_nrcodes[] = {0,0,1,5,1,1,1,1,1,1,0,0,0,0,0,0,0};
    static const unsigned char std_dc_luminance_values[] = {0,1,2,3,4,5,6,7,8,9,10,11};
@@ -1521,6 +1521,20 @@ static int stbi_write_jpg_core(stbi__write_context *s, int width, int height, in
       s->func(s->context, (void*)YTable, sizeof(YTable));
       stbiw__putc(s, 1);
       s->func(s->context, UVTable, sizeof(UVTable));
+
+      // comment block with parameters of generation
+      if(parameters != NULL) {
+         stbiw__putc(s, 0xFF /* comnent */ );
+         stbiw__putc(s, 0xFE /* marker  */ );
+         size_t param_length = std::min(2 + strlen("parameters") + 1 + strlen(parameters) + 1, (size_t) 0xFFFF);
+         stbiw__putc(s, param_length >> 8); // no need to mask, length < 65536
+         stbiw__putc(s, param_length & 0xFF);
+         s->func(s->context, (void*)"parameters", strlen("parameters") + 1); // std::string is zero-terminated
+         s->func(s->context, (void*)parameters, std::min(param_length, (size_t) 65534) - 2 - strlen("parameters") - 1);
+         if(param_length > 65534) stbiw__putc(s, 0); // always zero-terminate for safety
+         if(param_length & 1) stbiw__putc(s, 0xFF); // pad to even length
+      }
+
       s->func(s->context, (void*)head1, sizeof(head1));
       s->func(s->context, (void*)(std_dc_luminance_nrcodes+1), sizeof(std_dc_luminance_nrcodes)-1);
       s->func(s->context, (void*)std_dc_luminance_values, sizeof(std_dc_luminance_values));
@@ -1625,16 +1639,16 @@ STBIWDEF int stbi_write_jpg_to_func(stbi_write_func *func, void *context, int x,
 {
    stbi__write_context s = { 0 };
    stbi__start_write_callbacks(&s, func, context);
-   return stbi_write_jpg_core(&s, x, y, comp, (void *) data, quality);
+   return stbi_write_jpg_core(&s, x, y, comp, (void *) data, quality, NULL);
 }
 
 
 #ifndef STBI_WRITE_NO_STDIO
-STBIWDEF int stbi_write_jpg(char const *filename, int x, int y, int comp, const void *data, int quality)
+STBIWDEF int stbi_write_jpg(char const *filename, int x, int y, int comp, const void *data, int quality, const char* parameters)
 {
    stbi__write_context s = { 0 };
    if (stbi__start_write_file(&s,filename)) {
-      int r = stbi_write_jpg_core(&s, x, y, comp, data, quality);
+      int r = stbi_write_jpg_core(&s, x, y, comp, data, quality, parameters);
       stbi__end_write_file(&s);
       return r;
    } else

upscaler.cpp

@@ -15,7 +15,7 @@ struct UpscalerGGML {
     }
 
     bool load_from_file(const std::string& esrgan_path) {
-#ifdef SD_USE_CUBLAS
+#ifdef SD_USE_CUDA
         LOG_DEBUG("Using CUDA backend");
         backend = ggml_backend_cuda_init(0);
 #endif

util.cpp

@@ -348,7 +348,7 @@ void pretty_progress(int step, int steps, float time) {
         }
     }
     progress += "|";
-    printf(time > 1.0f ? "\r%s %i/%i - %.2fs/it" : "\r%s %i/%i - %.2fit/s",
+    printf(time > 1.0f ? "\r%s %i/%i - %.2fs/it" : "\r%s %i/%i - %.2fit/s\033[K",
            progress.c_str(), step, steps,
            time > 1.0f || time == 0 ? time : (1.0f / time));
     fflush(stdout);  // for linux