some minor fixes

Author: LostRuins

koboldcpp.py

@@ -1248,7 +1248,7 @@ def fetch_gpu_properties(testCL,testCU,testVK):
             for idx in range(0,4):
                 if(len(FetchedCUdevices)>idx):
                     if len(FetchedCUdeviceMem)>idx:
-                        dmem = int(FetchedCUdeviceMem[idx]) if AMDgpu else (int(FetchedCUdeviceMem[idx])*1024*1024)
+                        dmem = (int(FetchedCUdeviceMem[idx])*1024*1024) if AMDgpu else (int(FetchedCUdeviceMem[idx])*1024*1024)
                         lowestcumem = dmem if lowestcumem==0 else (dmem if dmem<lowestcumem else lowestcumem)
                     if len(FetchedCUfreeMem)>idx:
                         dmem = (int(FetchedCUfreeMem[idx])*1024*1024)

otherarch/tts_adapter.cpp

@@ -661,8 +661,8 @@ bool ttstype_load_model(const tts_load_model_inputs inputs)
         printf("\nLoading OuteTTS Model, OuteTTS: %s \nWavTokenizer: %s \n",modelfile_ttc.c_str(),modelfile_cts.c_str());
         if(modelfile_ttc=="" || modelfile_cts=="")
         {
-             printf("\nWarning: KCPP OuteTTS missing a file! Make sure both TTS and WavTokenizer models are loaded.\n");
-              return false;
+            printf("\nWarning: KCPP OuteTTS missing a file! Make sure both TTS and WavTokenizer models are loaded.\n");
+            return false;
         }
     }
 

otherarch/ttscpp/src/dac_model.cpp

@@ -123,11 +123,6 @@ static struct ggml_tensor * dac_build_audio_inputs(struct ggml_context * ctx, st
 
 struct dac_context * build_new_dac_context(struct dac_model * model, int n_threads, bool use_cpu) {
     dac_context * dctx = new dac_context(model, n_threads);
-    if (!use_cpu) {
-#ifdef GGML_USE_METAL
-        dctx->backend = ggml_backend_metal_init();
-#endif
-    }
     dctx->backend_cpu = ggml_backend_cpu_init();
     dctx->set_threads();
     dctx->build_schedule();

otherarch/ttscpp/src/kokoro_model.cpp

@@ -1485,11 +1485,7 @@ std::string get_espeak_id_from_kokoro_voice(std::string voice) {
 
 struct kokoro_duration_context * build_new_duration_kokoro_context(struct kokoro_model * model, int n_threads, bool use_cpu) {
     kokoro_duration_context * kctx = new kokoro_duration_context(model, n_threads);
-    if (!use_cpu) {
-#ifdef GGML_USE_METAL
-        kctx->backend = ggml_backend_metal_init();
-#endif
-    }
+
     kctx->backend_cpu = ggml_backend_cpu_init();
     kctx->set_threads();
     kctx->build_schedule();
@@ -1500,11 +1496,6 @@ struct kokoro_duration_context * build_new_duration_kokoro_context(struct kokoro
 
 struct kokoro_context * build_new_kokoro_context(struct kokoro_model * model, int n_threads, bool use_cpu) {
     kokoro_context * kctx = new kokoro_context(model, n_threads);
-    if (!use_cpu) {
-#ifdef GGML_USE_METAL
-        kctx->backend = ggml_backend_metal_init();
-#endif
-    }
     kctx->backend_cpu = ggml_backend_cpu_init();
     kctx->set_threads();
     kctx->build_schedule();

otherarch/ttscpp/src/kokoro_model.h

@@ -302,7 +302,7 @@ struct kokoro_model : tts_model {
     	compute_tensor_meta_cb = &fn;
         prep_constants(meta_ctx);
         prep_layers(meta_ctx);
-        tts_model::setup_from_file(meta_ctx, load_context, cpu_only, "kokoro", 1.6, post_load_tensor_bytes);
+        tts_model::setup_from_file(meta_ctx, load_context, cpu_only, "kokoro", 1.6, post_load_tensor_bytes + 4096); //add some extra space
     }
 };
 

otherarch/ttscpp/src/phonemizer.cpp

@@ -821,11 +821,11 @@ void populate_kokoro_ipa_map(std::string executable_path)
             }
         }
         myfile.close();
-		printf("\nPopulated Kokoro IPA: %d entries", kokoro_ipa_map.size());
+		printf("\nPopulated Kokoro IPA: %d entries\n", kokoro_ipa_map.size());
     }
     else
     {
-		printf("\nUnable to open Kokoro IPA file");
+		printf("\nUnable to open Kokoro IPA file\n");
     }
 }
 std::string found_word_to_ipa(std::string input)

otherarch/ttscpp/src/snac_model.cpp

@@ -110,11 +110,6 @@ static struct ggml_tensor * snac_build_audio_inputs(struct ggml_context * ctx, s
 
 snac_context * build_new_snac_context(struct snac_model * model, int n_threads, bool use_cpu) {
     snac_context * sctx = new snac_context(model, n_threads);
-    if (!use_cpu) {
-#ifdef GGML_USE_METAL
-        sctx->backend = ggml_backend_metal_init();
-#endif
-    }
     sctx->backend_cpu = ggml_backend_cpu_init();
     sctx->set_threads();
     sctx->build_schedule();

otherarch/ttscpp/src/tts_model.cpp

@@ -34,10 +34,7 @@ void runner_context::get_ggml_node_data(struct ggml_tensor * output_node, float
 
 void runner_context::set_threads() {
     if (backend != nullptr) {
-#ifdef GGML_USE_METAL
-        // this is form copied from llama.cpp, but has since been removed. I don't know if this should be tuned.
-        // ggml_backend_metal_set_n_cb(backend, 1);
-#endif
+
     }
     if (backend_cpu != nullptr) {
         ggml_backend_cpu_set_n_threads(backend_cpu, n_threads);
@@ -107,7 +104,7 @@ void tts_model::prep_buffers_and_context(bool cpu_only, float size_offset, uint3
     }
     size_t ctx_size = ggml_tensor_overhead() * (tensor_meta.n_tensors * size_offset);
     struct ggml_init_params params = {
-        /*.mem_size   =*/ ctx_size,
+        /*.mem_size   =*/ ctx_size + 4096,
         /*.mem_buffer =*/ NULL,
         /*.no_alloc   =*/ true,
     };

otherarch/ttscpp/src/ttst5_encoder_model.cpp

@@ -164,11 +164,6 @@ void t5_encoder::assign_weight(std::string name, ggml_tensor * tensor) {
 
 struct t5_context * build_new_t5_context(struct t5_encoder * model, int n_threads, bool use_cpu) {
 	t5_context * t5ctx = new t5_context(model, n_threads);
-    if (!use_cpu) {
-#ifdef GGML_USE_METAL
-        t5ctx->backend = ggml_backend_metal_init();
-#endif
-    }
     t5ctx->backend_cpu = ggml_backend_cpu_init();
     t5ctx->set_threads();
     t5ctx->build_schedule();