some tweaks and cleanup

Author: LostRuins

otherarch/tts_adapter.cpp

@@ -697,7 +697,7 @@ tts_generation_outputs ttstype_generate(const tts_generation_inputs inputs)
 
     if(!inputs.quiet)
     {
-        printf("\nTTS Generating (%d input tokens)...", prompt_inp.size());
+        printf("\nTTS Processing (%d input tokens)...\n", prompt_inp.size());
     }
 
     prompt_add(prompt_inp, model_ttc, "<|text_end|>\n<|audio_start|>\n", false, true);
@@ -771,6 +771,10 @@ tts_generation_outputs ttstype_generate(const tts_generation_inputs inputs)
             output.status = 0;
             return output;
         }
+        if(!inputs.quiet)
+        {
+            printf("\rTTS Generating (%d AudioTokens)", n_decode);
+        }
     }
 
     if(!inputs.quiet && ttsdebugmode==1)
@@ -818,7 +822,7 @@ tts_generation_outputs ttstype_generate(const tts_generation_inputs inputs)
             audio[i] = 0.0f;
         }
         //add some silence at the end
-        for (int i = 0; i < 24000/20; ++i) {
+        for (int i = 0; i < 24000/10; ++i) {
             audio.push_back(0.0f);
         }
 

src/llama-context.cpp

@@ -87,7 +87,7 @@ void llama_set_inputs(llama_context & lctx, const llama_ubatch & ubatch) {
         //GGML_ASSERT(lctx.inp_out_ids && "every model that can must skip unused outputs");
 
         if (!lctx.inp_out_ids) {
-            LLAMA_LOG_WARN("%s: 'lctx.inp_out_ids' is not created\n", __func__);
+            //LLAMA_LOG_WARN("%s: 'lctx.inp_out_ids' is not created\n", __func__);
         } else {
             const int64_t n_tokens = ubatch.n_tokens;
 

src/llama.cpp

@@ -11124,7 +11124,7 @@ static int llama_decode_impl(
 
     GGML_ASSERT_CONTINUE(n_tokens_all <= cparams.n_batch);
 
-    GGML_ASSERT_CONTINUE((cparams.causal_attn || cparams.n_ubatch >= n_tokens_all) && "non-causal attention requires n_ubatch >= n_tokens");
+    //GGML_ASSERT_CONTINUE((cparams.causal_attn || cparams.n_ubatch >= n_tokens_all) && "non-causal attention requires n_ubatch >= n_tokens");
 
     if (lctx.t_compute_start_us == 0) {
         lctx.t_compute_start_us = ggml_time_us();