Enhance user input handling for llama-run

The main motivation for this change is it was not handing ctrl-d
correctly. Modify `read_user_input` to handle EOF, "/bye" command,
and empty input cases. Introduce `get_user_input` function to
manage user input loop and handle different return cases.

Signed-off-by: Eric Curtin <[email protected]>

Author: ericcurtin

examples/run/run.cpp

@@ -801,7 +801,20 @@ static int generate(LlamaData & llama_data, const std::string & prompt, std::str
 
 static int read_user_input(std::string & user) {
     std::getline(std::cin, user);
-    return user.empty();  // Should have data in happy path
+    if (std::cin.eof()) {
+        printf("\n");
+        return 1;
+    }
+
+    if (user == "/bye") {
+        return 1;
+    }
+
+    if (user.empty()) {
+        return 2;
+    }
+
+    return 0;  // Should have data in happy path
 }
 
 // Function to generate a response based on the prompt
@@ -868,26 +881,45 @@ static bool is_stdout_a_terminal() {
 #endif
 }
 
-// Function to tokenize the prompt
+// Function to handle user input
+static int get_user_input(std::string & user_input, const std::string & user) {
+    while (true) {
+        const int ret = handle_user_input(user_input, user);
+        if (ret == 1) {
+            return 1;
+        }
+
+        if (ret == 2) {
+            continue;
+        }
+
+        break;
+    }
+
+    return 0;
+}
+
+// Main chat loop function
 static int chat_loop(LlamaData & llama_data, const std::string & user) {
     int prev_len = 0;
     llama_data.fmtted.resize(llama_n_ctx(llama_data.context.get()));
     static const bool stdout_a_terminal = is_stdout_a_terminal();
     while (true) {
-        // Get user input
         std::string user_input;
-        while (handle_user_input(user_input, user)) {
+        if (get_user_input(user_input, user) == 1) {
+            return 0;
         }
 
         add_message("user", user.empty() ? user_input : user, llama_data);
+
         int new_len;
-        if (apply_chat_template_with_error_handling(llama_data, true, new_len) < 0) {
+        if (apply_chat_template_with_error_handling(llama_data, true, new_len) == 1) {
             return 1;
         }
 
         std::string prompt(llama_data.fmtted.begin() + prev_len, llama_data.fmtted.begin() + new_len);
         std::string response;
-        if (generate_response(llama_data, prompt, response, stdout_a_terminal)) {
+        if (generate_response(llama_data, prompt, response, stdout_a_terminal) == 1) {
             return 1;
         }
 
@@ -896,7 +928,7 @@ static int chat_loop(LlamaData & llama_data, const std::string & user) {
         }
 
         add_message("assistant", response, llama_data);
-        if (apply_chat_template_with_error_handling(llama_data, false, prev_len) < 0) {
+        if (apply_chat_template_with_error_handling(llama_data, false, prev_len) == 1) {
             return 1;
         }
     }