[Bugfix] Use HF config fields as fallback when loading Mistral config (#29239)

Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
This commit is contained in:
Cyrus Leung
2025-11-23 02:22:48 +08:00
committed by GitHub
parent 730bd35378
commit d1cf8214e5
4 changed files with 25 additions and 4 deletions

View File

@@ -204,7 +204,19 @@ class MistralConfigParser(ConfigParserBase):
from vllm.transformers_utils.configs.mistral import adapt_config_dict
config = adapt_config_dict(config_dict)
# Get missing fields from HF config if available
try:
hf_config_dict, _ = PretrainedConfig.get_config_dict(
model,
revision=revision,
code_revision=code_revision,
token=_get_hf_token(),
**kwargs,
)
except OSError: # Not found
hf_config_dict = {}
config = adapt_config_dict(config_dict, defaults=hf_config_dict)
# Mistral configs may define sliding_window as list[int]. Convert it
# to int and add the layer_types list[str] to make it HF compatible