From ba6a97c390c36ba7258efeb9a387c5b2650780d5 Mon Sep 17 00:00:00 2001
From: Georgi Gerganov <ggerganov@gmail.com>
Date: Sat, 7 Sep 2024 21:32:43 +0300
Subject: [PATCH] common : do not add null tokens during warmup

ggml-ci
---
 common/common.cpp | 9 +++++++--
 1 file changed, 7 insertions(+), 2 deletions(-)

diff --git a/common/common.cpp b/common/common.cpp
index 639430131..015f56ae8 100644
--- a/common/common.cpp
+++ b/common/common.cpp
@@ -2514,10 +2514,15 @@ struct llama_init_result llama_init_from_gpt_params(gpt_params & params) {
         llama_token bos = llama_token_bos(model);
         llama_token eos = llama_token_eos(model);
         // some models (e.g. T5) don't have a BOS token
-        if (bos != -1) {
+        if (bos != LLAMA_TOKEN_NULL) {
             tmp.push_back(bos);
         }
-        tmp.push_back(eos);
+        if (eos != LLAMA_TOKEN_NULL) {
+            tmp.push_back(eos);
+        }
+        if (tmp.empty()) {
+            tmp.push_back(0);
+        }
 
         if (llama_model_has_encoder(model)) {
             llama_encode(lctx, llama_batch_get_one(tmp.data(), tmp.size(), 0, 0));