start prompt wasn't evald with default batch_size

bug fix
in versions 0.1.3-0.1.5 start prompt was not eval'd correctly with new default batch_size of 64.

examples/talk-llama/talk-llama.cpp

@@ -421,42 +421,6 @@ void allow_xtts_file(std::string path, int xtts_play_allowed) {
     }
 }
 
-// writes to temp file 0 or 1
-// @path: not used anymore, we store in temp
-// @xtts_play_allowed: 0=dont play xtts, 1=xtts can play
-/*
-void allow_xtts_file(std::string path, int xtts_play_allowed) {	
-    std::filesystem::path filePath = getTempDir() / "xtts_play_allowed.txt";
-
-    // Check if the file exists and read the current value
-    std::string currentValue;
-    if (std::filesystem::exists(filePath)) {
-        std::ifstream file(filePath);
-        if (file.is_open()) {
-            std::getline(file, currentValue);
-            file.close();
-        }
-    }
-	
-	std::string xtts_play_allowed_str = std::to_string(xtts_play_allowed);
-
-    // Only update the file if the value is different
-    if (currentValue != xtts_play_allowed_str) {
-        std::ofstream file(filePath, std::ios::trunc);
-        if (file.is_open()) {
-            file << xtts_play_allowed_str;
-            file.close();
-            return true;
-        } else {
-            printf("Unable to open file: %s\n", filePath)
-            return false;
-        }
-    }
-
-    return true;
-}
-*/
-
 // trim from start (in place)
 inline void ltrim(std::string &s) {
     s.erase(s.begin(), std::find_if(s.begin(), s.end(), [](unsigned char ch) {
@@ -1051,18 +1015,7 @@ int run(int argc, const char ** argv) {
         exit(0);
     }
 
-	//const std::string fileName{params.xtts_control_path};
-	//std::ifstream readStream{fileName};	
-	//if(!readStream.good()){
-	//	printf("Warning: %s file not found, xtts wont stop on user speech without it\n", params.xtts_control_path.c_str());
-	//	readStream.close();
-	//}
-	//else // control file is ok
-	//{
-	//	readStream.close();
-	//	allow_xtts_file(params.xtts_control_path, 1); // xtts can play
-	//}
-
+	allow_xtts_file(params.xtts_control_path, 1); // xtts can play	
 
     // whisper init
     struct whisper_context_params cparams = whisper_context_default_params();
@@ -1219,15 +1172,18 @@ int run(int argc, const char ** argv) {
 	float llama_start_time = get_current_time_ms();	
 
 	// NEW prompt eval
+	int n_past = 0;
 	// Calculate the number of chunks needed
 	size_t num_chunks = (embd_inp.size() + lcparams.n_batch - 1) / lcparams.n_batch;
 	// Iterate through the chunks and evaluate them
 	for (size_t i = 0; i < num_chunks; i++) {
 		// Calculate the start and end indices for the current chunk
 		size_t start_idx = i * lcparams.n_batch;
 		size_t end_idx = std::min((i + 1) * lcparams.n_batch, embd_inp.size());
+		size_t chunk_size = end_idx - start_idx;
 		// Evaluate the current chunk
-		llama_eval(ctx_llama, embd_inp.data() + start_idx, end_idx - start_idx, 0);
+		llama_eval(ctx_llama, embd_inp.data() + start_idx, chunk_size, n_past);
+		n_past += chunk_size;
 	}
 
 	float llama_end_time = get_current_time_ms();
@@ -1288,7 +1244,7 @@ int run(int argc, const char ** argv) {
     const int n_keep   = embd_inp.size();
     const int n_ctx    = llama_n_ctx(ctx_llama);
 
-    int n_past = n_keep;
+    n_past = n_keep;
     int n_prev = 64;
     std::vector<int> past_prev_arr{};
     int n_past_prev = 0; // token count that was before the last answer
@@ -1492,7 +1448,7 @@ int run(int argc, const char ** argv) {
                     } else {
                         text_heard += words[i] + " ";
                     }
-                }				
+                }					
 				if (params.print_energy) fprintf(stdout, " [text_heard: (%s)]\n", text_heard.c_str());				
 
                 // check if audio starts with the wake-up command if enabled
@@ -1712,7 +1668,7 @@ int run(int argc, const char ** argv) {
 							{
 								printf(" [Resetting context of %d tokens.]\n", embd_inp.size());						
 								embd_inp = ::llama_tokenize(ctx_llama, prompt_llama, true);
-								n_past = embd_inp.size();
+								n_past = 0;
 								// NEW prompt eval
 								// Calculate the number of chunks needed
 								size_t num_chunks = (embd_inp.size() + lcparams.n_batch - 1) / lcparams.n_batch;
@@ -1721,8 +1677,10 @@ int run(int argc, const char ** argv) {
 									// Calculate the start and end indices for the current chunk
 									size_t start_idx = i * lcparams.n_batch;
 									size_t end_idx = std::min((i + 1) * lcparams.n_batch, embd_inp.size());
+									size_t chunk_size = end_idx - start_idx;
 									// Evaluate the current chunk
-									llama_eval(ctx_llama, embd_inp.data() + start_idx, end_idx - start_idx, 0);
+									llama_eval(ctx_llama, embd_inp.data() + start_idx, chunk_size, n_past);
+									n_past += chunk_size;
 								}
 								printf(" [Context is now %d/%d tokens]\n", embd_inp.size(), params.ctx_size);		
 
@@ -1927,7 +1885,7 @@ int run(int argc, const char ** argv) {
 
 						// NEW prompt eval
 						size_t total_size = embd.size();
-						size_t batch_size = lcparams.n_batch;
+						//size_t batch_size = lcparams.n_batch;
 						// Split the input embeddings into smaller chunks
 						for (size_t i = 0; i < total_size; i += lcparams.n_batch) {
 							size_t chunk_size = (total_size - i < lcparams.n_batch) ? (total_size - i) : lcparams.n_batch;