[ci] bump transformers to 4.57.1 (#1586)

Signed-off-by: AlpinDale <alpindale@gmail.com>

Author: AlpinDale

aphrodite/modeling/models/moonvit.py

@@ -47,7 +47,7 @@
 import torch
 import torch.nn as nn
 import torch.nn.functional as F
-from transformers.activations import ACT2FN, PytorchGELUTanh
+from transformers.activations import ACT2FN
 from transformers.modeling_utils import PreTrainedModel
 from transformers.utils import is_flash_attn_2_available
 
@@ -596,7 +596,7 @@ def __init__(
                 "num_heads": config.num_attention_heads,
                 "hidden_dim": config.hidden_size,
                 "mlp_dim": config.intermediate_size,
-                "activation": PytorchGELUTanh(),
+                "activation": ACT2FN["gelu_pytorch_tanh"],
                 "attn_bias": True,
                 "attn_implementation": config._attn_implementation,
             },

aphrodite/modeling/models/qwen2_vl.py

@@ -31,7 +31,7 @@
 import torch.nn as nn
 import torch.nn.functional as F
 from einops import rearrange, repeat
-from transformers import AutoConfig, BatchFeature, PretrainedConfig
+from transformers import BatchFeature, PretrainedConfig
 from transformers.models.qwen2_vl import Qwen2VLImageProcessor, Qwen2VLProcessor
 from transformers.models.qwen2_vl.configuration_qwen2_vl import Qwen2VLConfig, Qwen2VLVisionConfig
 from transformers.models.qwen2_vl.image_processing_qwen2_vl import smart_resize
@@ -1500,9 +1500,7 @@ def __init__(
 class Tarsier2ProcessingInfo(Qwen2VLProcessingInfo):
     def get_hf_config(self) -> Qwen2VLConfig:
         model_path = self.ctx.model_config.model
-        original_config = AutoConfig.from_pretrained(model_path)
-        config_dict = original_config.to_dict()
-        correct_config = Qwen2VLConfig.from_dict(config_dict)
+        correct_config = Qwen2VLConfig.from_pretrained(model_path)
 
         return correct_config
 

aphrodite/modeling/models/transformers/base.py

@@ -293,7 +293,7 @@ def create_attention_instances(self) -> dict[int, Attention]:
         # Aphrodite does not support encoder-decoder models, so if any encoder layer is
         # found in a text only model, we assume the whole model is an encoder model
         if has_encoder(self.model) and not is_multimodal(self.config):
-            self.check_version("4.57.0.dev0", "encoder models support")
+            self.check_version("5.0.0", "encoder models support")
             attn_type = AttentionType.ENCODER_ONLY
         else:
             attn_type = AttentionType.DECODER

aphrodite/modeling/models/transformers/moe.py

@@ -112,7 +112,7 @@ def transformers_moe_forward_fake(
 
 class MoEMixin(MixtureOfExperts):
     def __init__(self, *, aphrodite_config: "AphroditeConfig", prefix: str = ""):
-        self.check_version("4.57.0.dev0", "MoE models support")
+        self.check_version("5.0.0", "MoE models support")
         # Skip MixtureOfExperts.__init__ and call the next class in MRO
         super(MixtureOfExperts, self).__init__(aphrodite_config=aphrodite_config, prefix=prefix)
 

requirements/common.txt

@@ -7,7 +7,7 @@ requests >= 2.26.0
 tqdm
 blake3
 py-cpuinfo
-transformers >= 4.56.0
+transformers >= 4.56.0, < 5
 tokenizers >= 0.21.1  # Required for fast incremental detokenization.
 protobuf # Required by LlamaTokenizer.
 fastapi[standard] >= 0.115.0 # Required by FastAPI's form models in the OpenAI API server's audio transcriptions endpoint.

requirements/test.in

@@ -37,7 +37,7 @@ datamodel_code_generator # required for minicpm3 test
 # TODO: Use lm-eval[api]==0.4.10 once released
 lm-eval[api] @ git+https://github.com/EleutherAI/lm-evaluation-harness.git@206b7722158f58c35b7ffcd53b035fdbdda5126d # required for model evaluation test
 mteb[bm25s]>=1.38.11, <2 # required for mteb test
-transformers==4.56.2
+transformers==4.57.1
 tokenizers==0.22.0
 schemathesis>=3.39.15 # Required for openai schema test.
 # quantization

requirements/test.txt

@@ -1196,7 +1196,7 @@ tqdm==4.66.6
     #   transformers
 tqdm-multiprocess==0.0.11
     # via lm-eval
-transformers==4.56.2
+transformers==4.57.1
     # via
     #   -r requirements/test.in
     #   genai-perf

tests/models/registry.py

@@ -745,21 +745,19 @@ def check_available_online(
 }
 
 _TRANSFORMERS_BACKEND_MODELS = {
-    "TransformersEmbeddingModel": _HfExamplesInfo("BAAI/bge-base-en-v1.5", min_transformers_version="4.57.0.dev0"),
+    "TransformersEmbeddingModel": _HfExamplesInfo("BAAI/bge-base-en-v1.5", min_transformers_version="5.0.0"),
     "TransformersForSequenceClassification": _HfExamplesInfo(
         "papluca/xlm-roberta-base-language-detection",
-        min_transformers_version="4.57.0.dev0",
+        min_transformers_version="5.0.0",
     ),
     "TransformersForCausalLM": _HfExamplesInfo("hmellor/Ilama-3.2-1B", trust_remote_code=True),
     "TransformersMultiModalForCausalLM": _HfExamplesInfo("BAAI/Emu3-Chat-hf"),
-    "TransformersMoEForCausalLM": _HfExamplesInfo("allenai/OLMoE-1B-7B-0924", min_transformers_version="4.57.0.dev0"),
+    "TransformersMoEForCausalLM": _HfExamplesInfo("allenai/OLMoE-1B-7B-0924", min_transformers_version="5.0.0"),
     "TransformersMultiModalMoEForCausalLM": _HfExamplesInfo(
-        "Qwen/Qwen3-VL-30B-A3B-Instruct", min_transformers_version="4.57.0.dev0"
-    ),
-    "TransformersMoEEmbeddingModel": _HfExamplesInfo("Qwen/Qwen3-30B-A3B", min_transformers_version="4.57.0.dev0"),
-    "TransformersMoEForSequenceClassification": _HfExamplesInfo(
-        "Qwen/Qwen3-30B-A3B", min_transformers_version="4.57.0.dev0"
+        "Qwen/Qwen3-VL-30B-A3B-Instruct", min_transformers_version="5.0.0"
     ),
+    "TransformersMoEEmbeddingModel": _HfExamplesInfo("Qwen/Qwen3-30B-A3B", min_transformers_version="5.0.0"),
+    "TransformersMoEForSequenceClassification": _HfExamplesInfo("Qwen/Qwen3-30B-A3B", min_transformers_version="5.0.0"),
     "TransformersMultiModalEmbeddingModel": _HfExamplesInfo("google/gemma-3-4b-it"),
     "TransformersMultiModalForSequenceClassification": _HfExamplesInfo("google/gemma-3-4b-it"),
 }

tests/models/test_transformers.py

@@ -80,7 +80,7 @@ def test_models(
     from packaging.version import Version
 
     installed = Version(transformers.__version__)
-    required = Version("4.57.0.dev0")
+    required = Version("5.0.0")
     if model == "allenai/OLMoE-1B-7B-0924" and installed < required:
         pytest.skip(f"MoE models with the Transformers backend require transformers>={required}, but got {installed}")