A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 13, 23:15 UTC
[CI/Build] Bump transformers version (#27528)
Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk> Signed-off-by: Isotr0py <mozf@mail2.sysu.edu.cn> Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com> Co-authored-by: Isotr0py <mozf@mail2.sysu.edu.cn> Co-authored-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
Cyrus Leung committed
Nov 1, 2025 at 13:11 UTC
879a06579ea8a057c250e9b6cc4e632dabd87d2e
9 files changed
+17
-17
requirements/common.txt
+1
-1
@@ -7,7 +7,7 @@ requests >= 2.26.0
7
tqdm
8
blake3
9
py-cpuinfo
10
-transformers >= 4.56.0
10
+transformers >= 4.56.0, < 5
11
tokenizers >= 0.21.1 # Required for fast incremental detokenization.
12
protobuf # Required by LlamaTokenizer.
13
fastapi[standard] >= 0.115.0 # Required by FastAPI's form models in the OpenAI API server's audio transcriptions endpoint.
requirements/nightly_torch_test.txt
+1
-1
@@ -29,7 +29,7 @@ opencv-python-headless >= 4.11.0 # required for video test
29
datamodel_code_generator # required for minicpm3 test
30
lm-eval[api] @ git+https://github.com/EleutherAI/lm-evaluation-harness.git@206b7722158f58c35b7ffcd53b035fdbdda5126d # required for model evaluation test
31
mteb>=1.38.11, <2 # required for mteb test
32
-transformers==4.56.2
32
+transformers==4.57.1
33
tokenizers==0.22.0
34
schemathesis>=3.39.15 # Required for openai schema test.
35
# quantization
requirements/test.in
+1
-1
@@ -37,7 +37,7 @@ datamodel_code_generator # required for minicpm3 test
37
# TODO: Use lm-eval[api]==0.4.10 once released
38
lm-eval[api] @ git+https://github.com/EleutherAI/lm-evaluation-harness.git@206b7722158f58c35b7ffcd53b035fdbdda5126d # required for model evaluation test
39
mteb[bm25s]>=1.38.11, <2 # required for mteb test
40
-transformers==4.56.2
40
+transformers==4.57.1
41
tokenizers==0.22.0
42
schemathesis>=3.39.15 # Required for openai schema test.
43
# quantization
requirements/test.txt
+1
-1
@@ -1196,7 +1196,7 @@ tqdm==4.66.6
1196
# transformers
1197
tqdm-multiprocess==0.0.11
1198
# via lm-eval
1199
-transformers==4.56.2
1199
+transformers==4.57.1
1200
# via
1201
# -r requirements/test.in
1202
# genai-perf
tests/models/multimodal/generation/test_maverick.py
+2
@@ -186,6 +186,8 @@ def create_reduced_config(
186
if "text_config" in config_dict:
187
original_text_layers = config_dict["text_config"]["num_hidden_layers"]
188
config_dict["text_config"]["num_hidden_layers"] = text_layers
189
+ original_layer_types = config_dict["text_config"]["layer_types"]
190
+ config_dict["text_config"]["layer_types"] = original_layer_types[:text_layers]
191
print(f"Reduced text layers from {original_text_layers} to {text_layers}")
192
193
original_num_experts = config_dict["text_config"]["num_local_experts"]
tests/models/registry.py
+6
-6
@@ -882,27 +882,27 @@ _SPECULATIVE_DECODING_EXAMPLE_MODELS = {
882
883
_TRANSFORMERS_BACKEND_MODELS = {
884
"TransformersEmbeddingModel": _HfExamplesInfo(
885
- "BAAI/bge-base-en-v1.5", min_transformers_version="4.57.0.dev0"
885
+ "BAAI/bge-base-en-v1.5", min_transformers_version="5.0.0"
886
),
887
"TransformersForSequenceClassification": _HfExamplesInfo(
888
"papluca/xlm-roberta-base-language-detection",
889
- min_transformers_version="4.57.0.dev0",
889
+ min_transformers_version="5.0.0",
890
),
891
"TransformersForCausalLM": _HfExamplesInfo(
892
"hmellor/Ilama-3.2-1B", trust_remote_code=True
893
),
894
"TransformersMultiModalForCausalLM": _HfExamplesInfo("BAAI/Emu3-Chat-hf"),
895
"TransformersMoEForCausalLM": _HfExamplesInfo(
896
- "allenai/OLMoE-1B-7B-0924", min_transformers_version="4.57.0.dev0"
896
+ "allenai/OLMoE-1B-7B-0924", min_transformers_version="5.0.0"
897
),
898
"TransformersMultiModalMoEForCausalLM": _HfExamplesInfo(
899
- "Qwen/Qwen3-VL-30B-A3B-Instruct", min_transformers_version="4.57.0.dev0"
899
+ "Qwen/Qwen3-VL-30B-A3B-Instruct", min_transformers_version="5.0.0"
900
),
901
"TransformersMoEEmbeddingModel": _HfExamplesInfo(
902
- "Qwen/Qwen3-30B-A3B", min_transformers_version="4.57.0.dev0"
902
+ "Qwen/Qwen3-30B-A3B", min_transformers_version="5.0.0"
903
),
904
"TransformersMoEForSequenceClassification": _HfExamplesInfo(
905
- "Qwen/Qwen3-30B-A3B", min_transformers_version="4.57.0.dev0"
905
+ "Qwen/Qwen3-30B-A3B", min_transformers_version="5.0.0"
906
),
907
"TransformersMultiModalEmbeddingModel": _HfExamplesInfo("google/gemma-3-4b-it"),
908
"TransformersMultiModalForSequenceClassification": _HfExamplesInfo(
tests/models/test_transformers.py
+1
-1
@@ -82,7 +82,7 @@ def test_models(
82
from packaging.version import Version
83
84
installed = Version(transformers.__version__)
85
- required = Version("4.57.0.dev0")
85
+ required = Version("5.0.0")
86
if model == "allenai/OLMoE-1B-7B-0924" and installed < required:
87
pytest.skip(
88
"MoE models with the Transformers backend require "
vllm/model_executor/models/moonvit.py
+2
-2
@@ -49,7 +49,7 @@ from functools import cached_property
49
import torch
50
import torch.nn as nn
51
import torch.nn.functional as F
52
-from transformers.activations import ACT2FN, PytorchGELUTanh
52
+from transformers.activations import ACT2FN
53
from transformers.modeling_utils import PreTrainedModel
54
from transformers.utils import is_flash_attn_2_available
55
@@ -651,7 +651,7 @@ class MoonVitPretrainedModel(PreTrainedModel):
651
"num_heads": config.num_attention_heads,
652
"hidden_dim": config.hidden_size,
653
"mlp_dim": config.intermediate_size,
654
- "activation": PytorchGELUTanh(),
654
+ "activation": ACT2FN["gelu_pytorch_tanh"],
655
"attn_bias": True,
656
"attn_implementation": config._attn_implementation,
657
},
vllm/model_executor/models/qwen2_vl.py
+2
-4
@@ -34,7 +34,7 @@ import torch
34
import torch.nn as nn
35
import torch.nn.functional as F
36
from einops import rearrange, repeat
37
-from transformers import AutoConfig, BatchFeature, PretrainedConfig
37
+from transformers import BatchFeature, PretrainedConfig
38
from transformers.models.qwen2_vl import Qwen2VLImageProcessor, Qwen2VLProcessor
39
from transformers.models.qwen2_vl.configuration_qwen2_vl import (
40
Qwen2VLConfig,
@@ -1651,9 +1651,7 @@ class Tarsier2Processor(Qwen2VLProcessor):
1651
class Tarsier2ProcessingInfo(Qwen2VLProcessingInfo):
1652
def get_hf_config(self) -> Qwen2VLConfig:
1653
model_path = self.ctx.model_config.model
1654
- original_config = AutoConfig.from_pretrained(model_path)
1655
- config_dict = original_config.to_dict()
1656
- correct_config = Qwen2VLConfig.from_dict(config_dict)
1654
+ correct_config = Qwen2VLConfig.from_pretrained(model_path)
1655
1656
return correct_config
1657