main: use dprintf and add --ctrl-token-no-out and --ctrl-token-fd-out

common/common.cpp

@@ -905,6 +905,14 @@ bool gpt_params_find_arg(int argc, char ** argv, const std::string & arg, gpt_pa
         params.interactive_specials = true;
         return true;
     }
+    if (arg == "--ctrl-token-no-out") {
+        params.ctrl_token_no_out = true;
+        return true;
+    }
+    if (arg == "--ctrl-token-fd-out") {
+        params.ctrl_token_fd_out = true;
+        return true;
+    }
     if (arg == "--embedding") {
         params.embedding = true;
         return true;
@@ -1433,7 +1441,10 @@ void gpt_print_usage(int /*argc*/, char ** argv, const gpt_params & params) {
     printf("  --version             show version and build info\n");
     printf("  -i, --interactive     run in interactive mode\n");
     printf("  --interactive-specials allow special tokens in user text, in interactive mode\n");
-    printf("  --interactive-first   run in interactive mode and wait for input right away\n");
+    printf("  --ctrl-token-no-out   control tokens output disabled\n");
+#if defined (__unix__) || (defined (__APPLE__) && defined (__MACH__))
+    printf("  --ctrl-token-fd-out   control tokens sent to file descriptor 3 out of band\n");
+#endif
     printf("  -cnv, --conversation  run in conversation mode (does not print special tokens and suffix/prefix)\n");
     printf("  -ins, --instruct      run in instruction mode (use with Alpaca models)\n");
     printf("  -cml, --chatml        run in chatml mode (use with ChatML-compatible models)\n");

common/common.h

@@ -142,6 +142,8 @@ struct gpt_params {
     bool use_color         = false; // use color to distinguish generations and inputs
     bool interactive       = false; // interactive mode
     bool interactive_specials = false; // whether to allow special tokens from user, during interactive mode
+    bool ctrl_token_no_out = false; // disable control token output
+    bool ctrl_token_fd_out = false; // enable control token output and redirect it to file descriptor 3
     bool conversation      = false; // conversation mode (does not print special tokens and suffix/prefix)
     bool chatml            = false; // chatml mode (used for models trained on chatml syntax)
     bool prompt_cache_all  = false; // save user input and generations to prompt cache

examples/main/main.cpp

@@ -18,6 +18,7 @@
 #if defined (__unix__) || (defined (__APPLE__) && defined (__MACH__))
 #include <signal.h>
 #include <unistd.h>
+#define CONTROL_TOKEN_FILE_DESCRIPTOR (3)
 #elif defined (_WIN32)
 #define WIN32_LEAN_AND_MEAN
 #ifndef NOMINMAX
@@ -528,21 +529,6 @@ int main(int argc, char ** argv) {
         exit(1);
     }
 
-    // Create the pipe for special token handling
-    int stok_pipe[2] = {0};
-    if (pipe(stok_pipe) == -1) {
-        fprintf(stderr, "%s: failed to initialize special token output stream\n", __func__);
-        exit(1);
-    }
-
-    close(stok_pipe[0]); // Read Special Token Not In Use
-
-    FILE *special_token_stream_output_fd = fdopen(stok_pipe[1], "w");
-    if (special_token_stream_output_fd == NULL) {
-        fprintf(stderr, "%s: failed to open special token output stream\n", __func__);
-        exit(1);
-    }
-
     while ((n_remain != 0 && !is_antiprompt) || params.interactive) {
         // predict
         if (!embd.empty()) {
@@ -758,12 +744,22 @@ int main(int argc, char ** argv) {
                 const std::string token_str = llama_token_to_piece(ctx, id);
 
                 // Console/Stream Output
-                if (llama_token_is_control_token(llama_get_model(ctx), id)) {
+                if (!llama_token_is_control_token(llama_get_model(ctx), id)) {
-                    // Stream Output Token To Special Token Output
-                    fprintf(special_token_stream_output_fd, "%s", token_str.c_str());
-                } else {
                     // Stream Output Token To Standard Output
                     fprintf(stdout, "%s", token_str.c_str());
+                } else if (!params.ctrl_token_no_out) {
+#if defined (__unix__) || (defined (__APPLE__) && defined (__MACH__))
+                    if (params.ctrl_token_fd_out) {
+                        // Stream Control Token To Special Token Output. Useful for debugging control token behaviour
+                        dprintf(CONTROL_TOKEN_FILE_DESCRIPTOR, "%s", token_str.c_str());
+                    }
+                    else
+#endif
+                    if (!params.conversation && sparams.grammar.empty())
+                    {
+                        // Stream Control Token To Standard Output as long as we are not in a conversation or grammar output
+                        fprintf(stdout, "%s", token_str.c_str());
+                    }
                 }
 
                 // Record Displayed Tokens To Log
@@ -983,8 +979,6 @@ int main(int argc, char ** argv) {
     llama_sampling_free(ctx_sampling);
     llama_backend_free();
 
-    fclose(special_token_stream_output_fd);
-
 #ifndef LOG_DISABLE_LOGS
     LOG_TEE("Log end\n");
 #endif // LOG_DISABLE_LOGS