[deps] bump transformers to 4.49.0 (#8564)

src/llamafactory/data/mm_plugin.py

@@ -27,6 +27,10 @@ from typing import TYPE_CHECKING, BinaryIO, Literal, Optional, TypedDict, Union
 import numpy as np
 import torch
 from transformers.image_utils import get_image_size, is_valid_image, to_numpy_array
+from transformers.models.mllama.processing_mllama import (
+    convert_sparse_cross_attention_mask_to_dense,
+    get_cross_attention_token_mask,
+)
 from typing_extensions import override
 
 from ..extras.constants import AUDIO_PLACEHOLDER, IGNORE_INDEX, IMAGE_PLACEHOLDER, VIDEO_PLACEHOLDER
@@ -51,17 +55,10 @@ if is_pyav_available():
     import av
 
 
-if is_transformers_version_greater_than("4.45.0"):
-    from transformers.models.mllama.processing_mllama import (
-        convert_sparse_cross_attention_mask_to_dense,
-        get_cross_attention_token_mask,
-    )
-
-
 if is_transformers_version_greater_than("4.52.0"):
     from transformers.image_utils import make_flat_list_of_images
     from transformers.video_utils import make_batched_videos
-elif is_transformers_version_greater_than("4.49.0"):
+else:
     from transformers.image_utils import make_batched_videos, make_flat_list_of_images
 
 

src/llamafactory/extras/misc.py

@@ -35,7 +35,6 @@ from transformers.utils import (
 from transformers.utils.versions import require_version
 
 from . import logging
-from .packages import is_transformers_version_greater_than
 
 
 _is_fp16_available = is_torch_npu_available() or is_torch_cuda_available()
@@ -94,15 +93,11 @@ def check_version(requirement: str, mandatory: bool = False) -> None:
 
 def check_dependencies() -> None:
     r"""Check the version of the required packages."""
-    check_version(
-        "transformers>=4.45.0,<=4.52.4,!=4.46.0,!=4.46.1,!=4.46.2,!=4.46.3,!=4.47.0,!=4.47.1,!=4.48.0,!=4.52.0"
-    )
+    check_version("transformers>=4.49.0,<=4.52.4,!=4.52.0")
     check_version("datasets>=2.16.0,<=3.6.0")
     check_version("accelerate>=1.3.0,<=1.7.0")
     check_version("peft>=0.14.0,<=0.15.2")
     check_version("trl>=0.8.6,<=0.9.6")
-    if is_transformers_version_greater_than("4.46.0") and not is_transformers_version_greater_than("4.48.1"):
-        logger.warning_rank0_once("There are known bugs in transformers v4.46.0-v4.48.0, please use other versions.")
 
 
 def calculate_tps(dataset: list[dict[str, Any]], metrics: dict[str, float], stage: Literal["sft", "rm"]) -> float:

src/llamafactory/hparams/parser.py

@@ -348,7 +348,7 @@ def get_train_args(args: Optional[Union[dict[str, Any], list[str]]] = None) -> _
     if finetuning_args.finetuning_type == "lora":
         # https://github.com/huggingface/transformers/blob/v4.50.0/src/transformers/trainer.py#L782
         training_args.label_names = training_args.label_names or ["labels"]
-    
+
     if "swanlab" in training_args.report_to and finetuning_args.use_swanlab:
         training_args.report_to.remove("swanlab")
 

src/llamafactory/model/loader.py

@@ -19,6 +19,7 @@ import torch
 from transformers import (
     AutoConfig,
     AutoModelForCausalLM,
+    AutoModelForImageTextToText,
     AutoModelForSeq2SeqLM,
     AutoModelForTextToWaveform,
     AutoModelForVision2Seq,
@@ -29,7 +30,6 @@ 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
@@ -39,10 +39,6 @@ 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
 
@@ -162,10 +158,7 @@ def load_model(
         else:
             if type(config) in AutoModelForVision2Seq._model_mapping.keys():  # image-text
                 load_class = AutoModelForVision2Seq
-            elif (
-                is_transformers_version_greater_than("4.46.0")
-                and type(config) in AutoModelForImageTextToText._model_mapping.keys()
-            ):  # image-text
+            elif 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