[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
+10 -10
View File
@@ -62,16 +62,16 @@ def test_nested_rope_validation_patch_preserves_flat_rope_parameters(monkeypatch
def original_validate_rope(config, *args, **kwargs):
calls.append(config)
from transformers import PretrainedConfig
from transformers import PreTrainedConfig
monkeypatch.setattr(PretrainedConfig, "validate_rope", original_validate_rope)
monkeypatch.setattr(PreTrainedConfig, "validate_rope", original_validate_rope)
_patch_hf_transformers_nested_rope_validation()
nested_rope_parameters = {
"full_attention": {"rope_type": "default"},
"original_max_position_embeddings": 32768,
}
PretrainedConfig.validate_rope(
PreTrainedConfig.validate_rope(
SimpleNamespace(rope_parameters=nested_rope_parameters)
)
assert nested_rope_parameters == {"full_attention": {"rope_type": "default"}}
@@ -81,7 +81,7 @@ def test_nested_rope_validation_patch_preserves_flat_rope_parameters(monkeypatch
"factor": 8.0,
"rope_theta": 500000.0,
}
PretrainedConfig.validate_rope(
PreTrainedConfig.validate_rope(
SimpleNamespace(rope_parameters=flat_rope_parameters)
)
assert flat_rope_parameters == {
@@ -730,9 +730,9 @@ def test_batch_invariant_breakable_cudagraph(
],
)
def test_dsa_models_select_matching_mtp(model_type, expected_architecture):
from transformers import PretrainedConfig
from transformers import PreTrainedConfig
hf_config = PretrainedConfig(
hf_config = PreTrainedConfig(
architectures=["DeepseekV32ForCausalLM"],
num_nextn_predict_layers=1,
)
@@ -2463,14 +2463,14 @@ def test_get_and_verify_max_len_with_nope_layers(
max_model_len, rope_parameters, expected_max_len
):
"""NoPE layers do not prevent deriving or scaling the context length."""
from transformers import PretrainedConfig
from transformers import PreTrainedConfig
from vllm.config.model import _get_and_verify_max_len
from vllm.transformers_utils.model_arch_config_convertor import (
ModelArchConfigConvertorBase,
)
hf_config = PretrainedConfig(
hf_config = PreTrainedConfig(
max_position_embeddings=4096,
original_max_position_embeddings=2048,
)
@@ -2514,14 +2514,14 @@ def test_get_and_verify_max_len_yarn_is_already_scaled(
for every YaRN variant, so scaling it again overstates the limit and lets
requests past the end of the cos/sin cache.
"""
from transformers import PretrainedConfig
from transformers import PreTrainedConfig
from vllm.config.model import _get_and_verify_max_len
from vllm.transformers_utils.model_arch_config_convertor import (
ModelArchConfigConvertorBase,
)
hf_config = PretrainedConfig(max_position_embeddings=32768)
hf_config = PreTrainedConfig(max_position_embeddings=32768)
hf_config.rope_parameters = {
"rope_type": rope_type,
"factor": factor,