Use model's chat template, BOS fix, delete stale test reports.

1 files changed, 7 insertions, 3 deletions

diff --git a/main.c b/main.c
index 266bc34..bac9c43 100644
--- a/main.c
+++ b/main.c
@@ -53,7 +53,11 @@ static void process_request(struct llama_model *model, const char *user_prompt)
 		{ "user",   user_prompt }
 	};
 
-	prompt_len = llama_chat_apply_template(NULL, messages, 2, true, NULL, 0);
+	const char *tmpl = llama_model_chat_template(model, NULL);
+	if (!tmpl)
+		fprintf(stderr, "Warning: model has no embedded chat template\n");
+
+	prompt_len = llama_chat_apply_template(tmpl, messages, 2, true, NULL, 0);
 	if (prompt_len <= 0) {
 		fprintf(stderr, "Error: failed to calculate chat template size\n");
 		return;
@@ -87,7 +91,7 @@ static void process_request(struct llama_model *model, const char *user_prompt)
 	}
 
 	n_prompt_tokens = -llama_tokenize(vocab, 
-		prompt, prompt_len, NULL, 0, true, true);
+		prompt, prompt_len, NULL, 0, false, true);
 
 	if (n_prompt_tokens <= 0) {
 		fprintf(stderr, "Error: tokenization sizing failed\n");
@@ -106,7 +110,7 @@ static void process_request(struct llama_model *model, const char *user_prompt)
 	prompt_tokens = MALLOC((size_t)n_prompt_tokens * sizeof(llama_token));
 
 	if (llama_tokenize(vocab, prompt, prompt_len, prompt_tokens, 
-		n_prompt_tokens, true, true) < 0) {
+		n_prompt_tokens, false, true) < 0) {
 		fprintf(stderr, "Error: Tokenization failed\n");
 		free(prompt);
 		free(prompt_tokens);