Merge branch 'hiyouga:main' into main

Former-commit-id: 4076f52c8ba7da4624a1fb3fa52a7170d1c3171e

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(

src/llamafactory/extras/misc.py

@@ -65,7 +65,7 @@ def check_dependencies() -> None:
         require_version("datasets>=2.14.3", "To fix: pip install datasets>=2.14.3")
         require_version("accelerate>=0.27.2", "To fix: pip install accelerate>=0.27.2")
         require_version("peft>=0.10.0", "To fix: pip install peft>=0.10.0")
-        require_version("trl>=0.8.1", "To fix: pip install trl>=0.8.1")
+        require_version("trl>=0.8.2", "To fix: pip install trl>=0.8.2")
 
 
 def count_parameters(model: torch.nn.Module) -> Tuple[int, int]: