[Refactor] Separate sequence and token pooling types (#32026)

Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>

tests/models/language/pooling/test_embedding.py

@@ -54,7 +54,7 @@ def test_models(
     vllm_extra_kwargs = {}
     if model == "ssmits/Qwen2-7B-Instruct-embed-base":
         vllm_extra_kwargs["pooler_config"] = PoolerConfig(
-            pooling_type="MEAN", normalize=False
+            seq_pooling_type="MEAN", normalize=False
         )
 
     max_model_len: int | None = 512

tests/models/language/pooling/test_mm_classifier_conversion.py

@@ -88,7 +88,7 @@ def test_gemma_multimodal(
         convert="classify",
         load_format="auto",
         hf_overrides=update_config,
-        pooler_config=PoolerConfig(pooling_type="LAST"),
+        pooler_config=PoolerConfig(seq_pooling_type="LAST"),
         max_model_len=512,
         enforce_eager=True,
         tensor_parallel_size=1,