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 9, 14:45 UTC
[Model] Bump transformers version for test registry (#33100)
Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
Cyrus Leung committed
Jan 27, 2026 at 02:53 UTC
c25dbee40dda0803e24482edc70f3c1bda39a060
6 files changed
+20
-20
tests/models/registry.py
+14
-14
@@ -256,7 +256,7 @@ _TEXT_GENERATION_EXAMPLE_MODELS = {
256
),
257
"Exaone4ForCausalLM": _HfExamplesInfo("LGAI-EXAONE/EXAONE-4.0-32B"),
258
"ExaoneMoEForCausalLM": _HfExamplesInfo(
259
- "LGAI-EXAONE/K-EXAONE-236B-A23B", min_transformers_version="5.0.0"
259
+ "LGAI-EXAONE/K-EXAONE-236B-A23B", min_transformers_version="5.1.0"
260
),
261
"Fairseq2LlamaForCausalLM": _HfExamplesInfo("mgleize/fairseq2-dummy-Llama-3.2-1B"),
262
"FalconForCausalLM": _HfExamplesInfo("tiiuae/falcon-7b"),
@@ -273,7 +273,7 @@ _TEXT_GENERATION_EXAMPLE_MODELS = {
273
"Glm4MoeForCausalLM": _HfExamplesInfo("zai-org/GLM-4.5"),
274
"Glm4MoeLiteForCausalLM": _HfExamplesInfo(
275
"zai-org/GLM-4.7-Flash",
276
- min_transformers_version="5.0.0.dev",
276
+ min_transformers_version="5.0.0",
277
),
278
"GPT2LMHeadModel": _HfExamplesInfo("openai-community/gpt2", {"alias": "gpt2"}),
279
"GPTBigCodeForCausalLM": _HfExamplesInfo(
@@ -650,7 +650,7 @@ _MULTIMODAL_EXAMPLE_MODELS = {
650
# [Decoder-only]
651
"AriaForConditionalGeneration": _HfExamplesInfo("rhymes-ai/Aria"),
652
"AudioFlamingo3ForConditionalGeneration": _HfExamplesInfo(
653
- "nvidia/audio-flamingo-3-hf", min_transformers_version="5.0.0.dev"
653
+ "nvidia/audio-flamingo-3-hf", min_transformers_version="5.0.0"
654
),
655
"AyaVisionForConditionalGeneration": _HfExamplesInfo("CohereLabs/aya-vision-8b"),
656
"BagelForConditionalGeneration": _HfExamplesInfo("ByteDance-Seed/BAGEL-7B-MoT"),
@@ -693,7 +693,7 @@ _MULTIMODAL_EXAMPLE_MODELS = {
693
"GlmAsrForConditionalGeneration": _HfExamplesInfo(
694
"zai-org/GLM-ASR-Nano-2512",
695
trust_remote_code=True,
696
- min_transformers_version="5.0",
696
+ min_transformers_version="5.0.0",
697
),
698
"GraniteVision": _HfExamplesInfo("ibm-granite/granite-vision-3.3-2b"),
699
"GraniteSpeechForConditionalGeneration": _HfExamplesInfo(
@@ -709,7 +709,7 @@ _MULTIMODAL_EXAMPLE_MODELS = {
709
"GlmOcrForConditionalGeneration": _HfExamplesInfo(
710
"zai-org/GLM-OCR",
711
is_available_online=False,
712
- min_transformers_version="5.0.0.dev",
712
+ min_transformers_version="5.1.0",
713
),
714
"H2OVLChatModel": _HfExamplesInfo(
715
"h2oai/h2ovl-mississippi-800m",
@@ -1048,7 +1048,7 @@ _SPECULATIVE_DECODING_EXAMPLE_MODELS = {
1048
"ExaoneMoeMTP": _HfExamplesInfo(
1049
"LGAI-EXAONE/K-EXAONE-236B-A23B",
1050
speculative_model="LGAI-EXAONE/K-EXAONE-236B-A23B",
1051
- min_transformers_version="5.0.0",
1051
+ min_transformers_version="5.1.0",
1052
),
1053
"Glm4MoeMTPModel": _HfExamplesInfo(
1054
"zai-org/GLM-4.5",
@@ -1057,13 +1057,13 @@ _SPECULATIVE_DECODING_EXAMPLE_MODELS = {
1057
"Glm4MoeLiteMTPModel": _HfExamplesInfo(
1058
"zai-org/GLM-4.7-Flash",
1059
speculative_model="zai-org/GLM-4.7-Flash",
1060
- min_transformers_version="5.0.0.dev",
1060
+ min_transformers_version="5.0.0",
1061
),
1062
"GlmOcrMTPModel": _HfExamplesInfo(
1063
"zai-org/GLM-OCR",
1064
speculative_model="zai-org/GLM-OCR",
1065
is_available_online=False,
1066
- min_transformers_version="5.0.0.dev",
1066
+ min_transformers_version="5.1.0",
1067
),
1068
"LongCatFlashMTPModel": _HfExamplesInfo(
1069
"meituan-longcat/LongCat-Flash-Chat",
@@ -1090,27 +1090,27 @@ _SPECULATIVE_DECODING_EXAMPLE_MODELS = {
1090
1091
_TRANSFORMERS_BACKEND_MODELS = {
1092
"TransformersEmbeddingModel": _HfExamplesInfo(
1093
- "BAAI/bge-base-en-v1.5", min_transformers_version="5.0.0.dev"
1093
+ "BAAI/bge-base-en-v1.5", min_transformers_version="5.0.0"
1094
),
1095
"TransformersForSequenceClassification": _HfExamplesInfo(
1096
"papluca/xlm-roberta-base-language-detection",
1097
- min_transformers_version="5.0.0.dev",
1097
+ min_transformers_version="5.0.0",
1098
),
1099
"TransformersForCausalLM": _HfExamplesInfo(
1100
"hmellor/Ilama-3.2-1B", trust_remote_code=True
1101
),
1102
"TransformersMultiModalForCausalLM": _HfExamplesInfo("BAAI/Emu3-Chat-hf"),
1103
"TransformersMoEForCausalLM": _HfExamplesInfo(
1104
- "allenai/OLMoE-1B-7B-0924", min_transformers_version="5.0.0.dev"
1104
+ "allenai/OLMoE-1B-7B-0924", min_transformers_version="5.0.0"
1105
),
1106
"TransformersMultiModalMoEForCausalLM": _HfExamplesInfo(
1107
- "Qwen/Qwen3-VL-30B-A3B-Instruct", min_transformers_version="5.0.0.dev"
1107
+ "Qwen/Qwen3-VL-30B-A3B-Instruct", min_transformers_version="5.0.0"
1108
),
1109
"TransformersMoEEmbeddingModel": _HfExamplesInfo(
1110
- "Qwen/Qwen3-30B-A3B", min_transformers_version="5.0.0.dev"
1110
+ "Qwen/Qwen3-30B-A3B", min_transformers_version="5.0.0"
1111
),
1112
"TransformersMoEForSequenceClassification": _HfExamplesInfo(
1113
- "Qwen/Qwen3-30B-A3B", min_transformers_version="5.0.0.dev"
1113
+ "Qwen/Qwen3-30B-A3B", min_transformers_version="5.0.0"
1114
),
1115
"TransformersMultiModalEmbeddingModel": _HfExamplesInfo("google/gemma-3-4b-it"),
1116
"TransformersMultiModalForSequenceClassification": _HfExamplesInfo(
tests/models/test_transformers.py
+1
-1
@@ -78,7 +78,7 @@ def test_models(
78
from packaging.version import Version
79
80
installed = Version(transformers.__version__)
81
- required = Version("5.0.0.dev")
81
+ required = Version("5.0.0")
82
if model == "allenai/OLMoE-1B-7B-0924" and installed < required:
83
pytest.skip(
84
"MoE models with the Transformers modeling backend require "
tests/v1/e2e/test_spec_decode.py
+1
-1
@@ -455,7 +455,7 @@ def test_eagle_correctness(
455
from packaging.version import Version
456
457
installed = Version(transformers.__version__)
458
- required = Version("5.0.0.dev")
458
+ required = Version("5.0.0")
459
if installed < required:
460
pytest.skip(
461
"Eagle3 with the Transformers modeling backend requires "
vllm/model_executor/models/transformers/base.py
+2
-2
@@ -350,7 +350,7 @@ class Base(
350
# vLLM does not support encoder-decoder models, so if any encoder layer is
351
# found in a text only model, we assume the whole model is an encoder model
352
if has_encoder(self.model) and not is_multimodal(self.config):
353
- self.check_version("5.0.0.dev0", "encoder models support")
353
+ self.check_version("5.0.0", "encoder models support")
354
attn_type = AttentionType.ENCODER_ONLY
355
else:
356
attn_type = AttentionType.DECODER
@@ -502,7 +502,7 @@ class Base(
502
)
503
504
def set_aux_hidden_state_layers(self, layers: tuple[int, ...]) -> None:
505
- self.check_version("5.0.0.dev0", "Eagle3 support")
505
+ self.check_version("5.0.0", "Eagle3 support")
506
from transformers.utils.generic import OutputRecorder
507
508
# The default value in PreTrainedModel is None
vllm/model_executor/models/transformers/moe.py
+1
-1
@@ -118,7 +118,7 @@ direct_register_custom_op(
118
119
class MoEMixin(MixtureOfExperts):
120
def __init__(self, *, vllm_config: "VllmConfig", prefix: str = ""):
121
- self.check_version("5.0.0.dev0", "MoE models support")
121
+ self.check_version("5.0.0", "MoE models support")
122
# Skip MixtureOfExperts.__init__ and call the next class in MRO
123
super(MixtureOfExperts, self).__init__(vllm_config=vllm_config, prefix=prefix)
124
vllm/transformers_utils/config.py
+1
-1
@@ -328,7 +328,7 @@ def patch_rope_parameters(config: PretrainedConfig) -> None:
328
partial_rotary_factor = getattr_iter(config, names, None, warn=True)
329
ompe = getattr(config, "original_max_position_embeddings", None)
330
331
- if Version(version("transformers")) < Version("5.0.0.dev0"):
331
+ if Version(version("transformers")) < Version("5.0.0"):
332
# Transformers v4 installed, legacy config fields may be present
333
if (rope_scaling := getattr(config, "rope_scaling", None)) is not None:
334
config.rope_parameters = rope_scaling