[Frontend][4/n] Improve pooling entrypoints | pooling. (#39153)

Signed-off-by: wang.yuqi <yuqi.wang@daocloud.io>
This commit is contained in:
wang.yuqi
2026-04-09 10:09:45 +00:00
committed by GitHub
parent b6c9be509e
commit 66c079ae83
43 changed files with 554 additions and 733 deletions
+17
View File
@@ -12,6 +12,23 @@ from vllm.utils.import_utils import resolve_obj_by_qualname
logger = logging.getLogger(__name__)
def has_io_processor(
vllm_config: VllmConfig,
plugin_from_init: str | None = None,
):
if plugin_from_init:
model_plugin = plugin_from_init
else:
# A plugin can be specified via the model config
# Retrieve the model specific plugin if available
# This is using a custom field in the hf_config for the model
hf_config = vllm_config.model_config.hf_config.to_dict()
config_plugin = hf_config.get("io_processor_plugin")
model_plugin = config_plugin
return model_plugin is not None
def get_io_processor(
vllm_config: VllmConfig,
renderer: BaseRenderer,