Preserve revision pins in secondary artifact loaders (#51446)

Signed-off-by: Clinton Thomas <1033162+KernelClint@users.noreply.github.com>
Co-authored-by: Lucas Bourtoule <35483370+dhalf@users.noreply.github.com>
This commit is contained in:
Clinton Thomas
2026-08-11 04:33:10 +00:00
committed by GitHub
co-authored by Lucas Bourtoule
parent adc1200597
commit 90fd4a333f
11 changed files with 226 additions and 13 deletions
+47
View File
@@ -6,7 +6,15 @@ only get the `eos_token_id` from the tokenizer as defined by
`BaseRenderer.get_eos_token_id`.
"""
from types import SimpleNamespace
from typing import cast
from unittest.mock import patch
from transformers import PretrainedConfig
from vllm.config.model import ModelConfig
from vllm.tokenizers import get_tokenizer
from vllm.transformers_utils import config as config_module
from vllm.transformers_utils.config import try_get_generation_config
@@ -30,3 +38,42 @@ def test_get_blip2_eos_token():
generation_config = try_get_generation_config(model_name, trust_remote_code=False)
assert generation_config is not None
assert generation_config.eos_token_id == 50118
def test_model_config_generation_fallback_forwards_code_revision():
model_config = cast(
ModelConfig,
SimpleNamespace(
generation_config="auto",
hf_config_path=None,
model="org/model",
trust_remote_code=True,
revision="model-pin",
code_revision="code-pin",
config_format="auto",
hf_token=None,
),
)
with (
patch.object(
config_module.GenerationConfig,
"from_pretrained",
side_effect=OSError,
),
patch.object(
config_module,
"get_config",
return_value=PretrainedConfig(),
) as get_config,
):
ModelConfig.try_get_generation_config(model_config)
get_config.assert_called_once_with(
"org/model",
trust_remote_code=True,
revision="model-pin",
code_revision="code-pin",
config_format="auto",
token=None,
)