From 7134fb02bbdc9421f6c314ae176d5786a8cd768d Mon Sep 17 00:00:00 2001
From: hiyouga <467089858@qq.com>
Date: Tue, 21 May 2024 20:03:09 +0800
Subject: [PATCH] fix paligemma sft

---
 src/llamafactory/data/preprocess.py | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/src/llamafactory/data/preprocess.py b/src/llamafactory/data/preprocess.py
index 4bc5ad3c..7bf9d4bc 100644
--- a/src/llamafactory/data/preprocess.py
+++ b/src/llamafactory/data/preprocess.py
@@ -89,7 +89,7 @@ def preprocess_supervised_dataset(
         if processor is not None and hasattr(processor, "image_seq_length"):  # paligemma case
             image_token_id = tokenizer.convert_tokens_to_ids(IMAGE_TOKEN)
             input_ids += [image_token_id] * getattr(processor, "image_seq_length")
-            labels += [image_token_id] * getattr(processor, "image_seq_length")
+            labels += [IGNORE_INDEX] * getattr(processor, "image_seq_length")
 
         for turn_idx, (source_ids, target_ids) in enumerate(
             template.encode_multiturn(