Fix OBSOLETE hallucination.

1 files changed, 29 insertions, 18 deletions

diff --git a/main.c b/main.c
index 602a8fd..42db9c6 100644
--- a/main.c
+++ b/main.c
@@ -1,16 +1,16 @@
 #include <stdio.h>
 #include <stdbool.h>
+#include <ctype.h>
 #include <err.h>
 #include <unistd.h>
 
-#include "llama.h"
 #include "mem.h"
+#include "llama.h"
 
 #define MODEL_PATH "Qwen2.5-7B-Instruct-Q4_K_M.gguf"
 
 #define SYSTEM_PROMPT \
-	"You are a precise lexicographer. You are defining the EXACT target word provided.\n" \
-	"Do NOT confuse the word with phonetically or visually similar words.\n\n" \
+	"You are a lexicographer. You are defining the EXACT target word provided.\n" \
 	"Output format strictly as follows:\n" \
 	"<WORD IN UPPERCASE> (<part of speech>) — <Formality: Conversational|Formal|Literary|Archaic>\n\n" \
 	"DEFINITION:\n" \
@@ -25,32 +25,37 @@
 #define CHATML_FMT \
 	"<|im_start|>system\n%s<|im_end|>\n" \
 	"<|im_start|>user\nTarget word: %s\nDefine this specific word:<|im_end|>\n" \
-	"<|im_start|>assistant\n"
+	"<|im_start|>assistant\n%s"
 
 #define MAX_TOKENS 300
 
+static void upcase(char *dest, const char *src, size_t n)
+{
+	size_t i;
+
+	for (i = 0; i < n - 1 && src[i] != '\0'; i++)
+		dest[i] = (char)toupper((unsigned char)src[i]);
+
+	dest[i] = '\0';
+}
+
 static int build_prompt(char **buf, const char *word)
 {
-	char *p;
-	int len, written;
+	int len;
+	char upper_word[128];
 
 	if (!buf || !word)
 		return -1;
 
-	len = snprintf(NULL, 0, CHATML_FMT, SYSTEM_PROMPT, word);
-	if (len <= 0)
-		return -1;
-
-	p = MALLOC((size_t)len + 1);
+	upcase(upper_word, word, sizeof(upper_word));
 
-	written = snprintf(p, (size_t)len + 1, CHATML_FMT, SYSTEM_PROMPT, word);
-	if (written <= 0) {
-		free(p);
+	len = asprintf(buf, CHATML_FMT, SYSTEM_PROMPT, word, upper_word);
+	if (len < 0) {
+		*buf = NULL;
 		return -1;
 	}
-	
-	*buf = p;
-	return written;
+
+	return len;
 }
 
 static void process_request(struct llama_model *model, const char *word)
@@ -141,6 +146,12 @@ static void process_request(struct llama_model *model, const char *word)
 
 	free(prompt_tokens);
 
+	/* Response formatting: print the pre-filled word right as generation begins */
+	char upper_word[128];
+	upcase(upper_word, word, sizeof(upper_word));
+	printf("%s", upper_word);
+	fflush(stdout);
+
 	/* Generation loop */
 	sparams = llama_sampler_chain_default_params();
 	smpl = llama_sampler_chain_init(sparams);
@@ -201,7 +212,7 @@ static void process_request(struct llama_model *model, const char *word)
 		}
 	}
 
-	printf("\n");
+	printf("\n\n");
 	fflush(stdout);
 
 	llama_perf_context_print(ctx);