[breaking] bump transformers to 4.45.0 & improve ci (#7746)

* update ci

* fix

* fix

* fix

* fix

* fix

src/llamafactory/model/loader.py

@@ -19,7 +19,6 @@ import torch
 from transformers import (
     AutoConfig,
     AutoModelForCausalLM,
-    AutoModelForImageTextToText,
     AutoModelForSeq2SeqLM,
     AutoModelForTextToWaveform,
     AutoModelForVision2Seq,
@@ -30,6 +29,7 @@ from trl import AutoModelForCausalLMWithValueHead
 
 from ..extras import logging
 from ..extras.misc import count_parameters, skip_check_imports, try_download_model_from_other_hub
+from ..extras.packages import is_transformers_version_greater_than
 from .adapter import init_adapter
 from .model_utils.liger_kernel import apply_liger_kernel
 from .model_utils.misc import register_autoclass
@@ -39,6 +39,10 @@ from .model_utils.valuehead import load_valuehead_params
 from .patcher import patch_config, patch_model, patch_processor, patch_tokenizer, patch_valuehead_model
 
 
+if is_transformers_version_greater_than("4.46.0"):
+    from transformers import AutoModelForImageTextToText
+
+
 if TYPE_CHECKING:
     from transformers import PretrainedConfig, PreTrainedModel, PreTrainedTokenizer, ProcessorMixin
 
@@ -145,7 +149,10 @@ def load_model(
         else:
             if type(config) in AutoModelForVision2Seq._model_mapping.keys():  # image-text
                 load_class = AutoModelForVision2Seq
-            elif type(config) in AutoModelForImageTextToText._model_mapping.keys():  # image-text
+            elif (
+                is_transformers_version_greater_than("4.46.0")
+                and type(config) in AutoModelForImageTextToText._model_mapping.keys()
+            ):  # image-text
                 load_class = AutoModelForImageTextToText
             elif type(config) in AutoModelForSeq2SeqLM._model_mapping.keys():  # audio-text
                 load_class = AutoModelForSeq2SeqLM