[FIX] Throwing an exception when the model does not support pool tasks (#25840)

[yyzxw]

Purpose

fix #25840

Test Plan

Test Result

---

Essential Elements of an Effective PR Description Checklist

• The purpose of the PR, such as "Fix some issue (link existing issues this PR will resolve)".
• The test plan, such as providing test command.
• The test results, such as pasting the results comparison before and after, or e2e results
• (Optional) The necessary documentation update, such as updating supported_models.md and examples for a new model.
• (Optional) Release notes update. If your change is user facing, please update the release notes draft in the Google Doc.

[gemini-code-assist]

Code Review

This pull request correctly addresses a ValueError that occurs when _dummy_pooler_run is called for a model that does not support any pooling tasks. The fix introduces a check at the beginning of the function to raise a RuntimeError with a clear message, which is a good improvement. I have a couple of suggestions to further refine the code by removing a redundant function call and a now-unnecessary assertion.

[noooop]

Sorry for overlooking as_reward_model in #20930

try add @default_pooling_type("ALL") before L390

vllm/vllm/model_executor/models/adapters.py

Lines 387 to 392 in bd51f78

	# Lazy import
	from vllm.model_executor.layers.pooler import DispatchPooler, Pooler
	class ModelForReward(_create_pooling_model_cls(cls)):
	def _init_pooler(self, vllm_config: "VllmConfig", prefix: str = ""):

And comment: The Reward model uses all pooling by default

refer to

vllm/vllm/model_executor/models/qwen2_rm.py

Lines 96 to 100 in bd51f78

	@default_pooling_type("ALL")
	class Qwen2ForRewardModel(Qwen2RewardBaseModel):
	def __init__(self, *, vllm_config: VllmConfig, prefix: str = ""):

[yyzxw]

@noooop why should add @default_pooling_type("ALL") to /vllm/model_executor/models/adapters.py#L390？
I tried to add it, but nothing changed，still get error.

[noooop]

@noooop why should add @default_pooling_type("ALL") to /vllm/model_executor/models/adapters.py#L390？ I tried to add it, but nothing changed，still get error.

vllm/vllm/config/__init__.py

Lines 407 to 418 in 9360d34

	if self.model_config:
	if self.model_config.pooler_config:
	pooling_type = self.model_config.pooler_config.pooling_type
	if pooling_type is None or pooling_type.lower() != "last":
	disable_chunked_prefill_reasons.append(
	"Only \"last\" pooling supports chunked "
	"prefill and prefix caching; disabling both.")
	if not getattr(self.model_config.hf_config, "is_causal", True):
	disable_chunked_prefill_reasons.append(
	"Only models using causal attention supports chunked "
	"prefill and prefix caching; disabling both.")
	elif self.model_config.is_encoder_decoder:

L410 requires pooling_type.lower() != "last" anyway. This will automatically turn off chunked prefill.

Please find a way to make setting @default_pooling_type("ALL") work

[noooop]

@yyzxw Thanks for your contribution

language-models-test-extended-pooling failure is unrelated to this PR

Will be fixed in #25817

@DarkLight1337 Is there anything else that needs to be modified in this PR?

[DarkLight1337]

No