[Chore] Use Transformers v5 names and drop redundant processor use_fast (#58550)

Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
Co-authored-by: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
This commit is contained in:
Harry Mellor
2026-09-24 14:14:10 +00:00
committed by GitHub
co-authored by Claude Opus 5.5
parent 484c211fe8
commit d881812500
218 changed files with 869 additions and 813 deletions
+3 -3
View File
@@ -3,7 +3,7 @@
from typing import get_args
from transformers import PretrainedConfig
from transformers import PreTrainedConfig
from vllm.config.speculative import MTPModelTypes, SpeculativeConfig
from vllm.transformers_utils.model_arch_config_convertor import (
@@ -11,8 +11,8 @@ from vllm.transformers_utils.model_arch_config_convertor import (
)
def _bailing_config() -> PretrainedConfig:
config = PretrainedConfig(
def _bailing_config() -> PreTrainedConfig:
config = PreTrainedConfig(
architectures=["BailingMoeV2_5ForCausalLM"],
hidden_size=4096,
kv_lora_rank=512,
+8 -8
View File
@@ -10,7 +10,7 @@ from typing import cast
from unittest.mock import Mock
import pytest
from transformers import PretrainedConfig
from transformers import PreTrainedConfig
from transformers.models.gemma4.configuration_gemma4 import Gemma4TextConfig
from vllm.config import ModelConfig, ParallelConfig, SpeculativeConfig
@@ -125,7 +125,7 @@ def _assert_model_config_methods(
def test_head_size_falls_back_when_head_dim_is_zero():
"""Regression test for configs that materialize missing head_dim as 0."""
hf_config = PretrainedConfig(
hf_config = PreTrainedConfig(
model_type="deepseek_vl_v2",
hidden_size=1280,
num_attention_heads=10,
@@ -140,7 +140,7 @@ def test_head_size_falls_back_when_head_dim_is_zero():
def test_qk_rope_head_dim_recovery_uses_model_revision(monkeypatch: pytest.MonkeyPatch):
hf_config = PretrainedConfig(
hf_config = PreTrainedConfig(
model_type="deepseek_v2",
qk_rope_head_dim=128,
qk_nope_head_dim=128,
@@ -176,7 +176,7 @@ def test_qk_rope_head_dim_recovery_uses_model_revision(monkeypatch: pytest.Monke
],
)
def test_num_experts_per_tok_aliases(attribute: str):
hf_config = PretrainedConfig(**{attribute: 4})
hf_config = PreTrainedConfig(**{attribute: 4})
convertor = ModelArchConfigConvertorBase(hf_config, hf_config)
model_config = cast(
ModelConfig,
@@ -191,7 +191,7 @@ def test_num_experts_per_tok_aliases(attribute: str):
def test_num_experts_per_tok_none_is_normalized():
hf_config = PretrainedConfig(top_k_experts=None)
hf_config = PreTrainedConfig(top_k_experts=None)
convertor = ModelArchConfigConvertorBase(hf_config, hf_config)
assert convertor.get_num_experts_per_token() == 0
@@ -207,7 +207,7 @@ def test_legacy_modelopt_config_without_producer_is_normalized():
"modelopt_quant_config": {"quant_cfg": {}},
}
}
hf_config = PretrainedConfig(quantization_config=quantization_config)
hf_config = PreTrainedConfig(quantization_config=quantization_config)
convertor = ModelArchConfigConvertorBase(hf_config, hf_config)
@@ -370,7 +370,7 @@ def test_gemma4_uniform_head_dims_are_homogeneous():
assert arch[3] is arch
class _HeterogeneousConfig(PretrainedConfig):
class _HeterogeneousConfig(PreTrainedConfig):
"""A heterogeneous config with no convertor of its own.
Mirrors the parts vLLM uses: per-layer configs are shallow copies with the
@@ -384,7 +384,7 @@ class _HeterogeneousConfig(PretrainedConfig):
super().__init__(**kwargs)
@property
def per_layer_config(self) -> list[PretrainedConfig]:
def per_layer_config(self) -> list[PreTrainedConfig]:
layers = []
for i in range(self.num_hidden_layers):
layer = copy(self)
+2 -2
View File
@@ -6,7 +6,7 @@ from unittest.mock import MagicMock, patch
import pytest
import torch
from transformers import PretrainedConfig
from transformers import PreTrainedConfig
from tests.models.utils import build_model_context
from vllm.config.ec_transfer import ECRole, ECTransferConfig
@@ -136,7 +136,7 @@ def test_language_model_only_disables_via_supports_multimodal_inputs():
def test_convertor_clears_mm_prefix_when_multimodal_disabled():
hf_config = PretrainedConfig(
hf_config = PreTrainedConfig(
model_type="gemma3",
architectures=["Gemma3ForConditionalGeneration"],
)
@@ -15,20 +15,20 @@ import functools
from unittest.mock import MagicMock, patch
import pytest
from transformers import PretrainedConfig
from transformers import PreTrainedConfig
from vllm.config.parallel import ParallelConfig
from vllm.config.speculative import SpeculativeConfig
def _make_hf_config(**kwargs) -> PretrainedConfig:
def _make_hf_config(**kwargs) -> PreTrainedConfig:
defaults = dict(
architectures=["LlamaForCausalLM"],
model_type="llama",
num_hidden_layers=64,
)
defaults.update(kwargs)
return PretrainedConfig(**defaults)
return PreTrainedConfig(**defaults)
@pytest.mark.cpu_test
@@ -137,7 +137,7 @@ def test_callable_overrides_reach_the_draft_config():
"""A callable override (config-to-config transform) composes with the
architecture-mapping override and is applied to the draft config."""
def shrink(hf_config: PretrainedConfig) -> PretrainedConfig:
def shrink(hf_config: PreTrainedConfig) -> PreTrainedConfig:
hf_config.num_hidden_layers = 1
return hf_config
@@ -155,7 +155,7 @@ def test_arch_mapping_applies_before_callable_override():
observes (and may adjust) the post-mapping config."""
seen_architectures: list[str] = []
def record(hf_config: PretrainedConfig) -> PretrainedConfig:
def record(hf_config: PreTrainedConfig) -> PreTrainedConfig:
seen_architectures.append(hf_config.architectures[0])
return hf_config
@@ -202,7 +202,7 @@ def test_inkling_override_exposes_all_mtp_depths():
assert out.local_layer_ids == [0, 2, 4]
def _module_level_shrink(hf_config: PretrainedConfig) -> PretrainedConfig:
def _module_level_shrink(hf_config: PreTrainedConfig) -> PreTrainedConfig:
hf_config.num_hidden_layers = 1
return hf_config
@@ -17,7 +17,7 @@ import logging
from typing import Literal
import pytest
from transformers import PretrainedConfig
from transformers import PreTrainedConfig
from vllm.config.model import ModelConfig
from vllm.config.parallel import ParallelConfig
@@ -46,7 +46,7 @@ def _override_logged(caplog: pytest.LogCaptureFixture) -> bool:
@pytest.mark.cpu_test
def test_override_raises_smaller_value(vllm_caplog: pytest.LogCaptureFixture):
hf_config = PretrainedConfig(max_position_embeddings=2048)
hf_config = PreTrainedConfig(max_position_embeddings=2048)
SpeculativeConfig._maybe_override_draft_max_position_embeddings(
hf_config, target_max_model_len=8192
)
@@ -56,7 +56,7 @@ def test_override_raises_smaller_value(vllm_caplog: pytest.LogCaptureFixture):
@pytest.mark.cpu_test
def test_override_keeps_sufficient_value(vllm_caplog: pytest.LogCaptureFixture):
hf_config = PretrainedConfig(max_position_embeddings=8192)
hf_config = PreTrainedConfig(max_position_embeddings=8192)
SpeculativeConfig._maybe_override_draft_max_position_embeddings(
hf_config, target_max_model_len=8192
)
@@ -66,7 +66,7 @@ def test_override_keeps_sufficient_value(vllm_caplog: pytest.LogCaptureFixture):
@pytest.mark.cpu_test
def test_override_ignores_missing_attribute(vllm_caplog: pytest.LogCaptureFixture):
hf_config = PretrainedConfig()
hf_config = PreTrainedConfig()
hf_config.__dict__.pop("max_position_embeddings", None)
SpeculativeConfig._maybe_override_draft_max_position_embeddings(
hf_config, target_max_model_len=8192