[Model] Re-add the implicit conversion feature for as_seq_cls_model (#21103)

Signed-off-by: wang.yuqi <noooop@126.com>

tests/models/test_transformers.py

@@ -138,3 +138,38 @@ def test_quantization(
         name_0="transformers",
         name_1="vllm",
     )
+
+
+@pytest.mark.parametrize(
+    "model",
+    ["jason9693/Qwen2.5-1.5B-apeach"],
+)
+@pytest.mark.parametrize("dtype", ["half"])
+def test_classify(
+    hf_runner,
+    vllm_runner,
+    example_prompts,
+    model: str,
+    dtype: str,
+    monkeypatch,
+) -> None:
+    import torch
+    from transformers import AutoModelForSequenceClassification
+
+    with vllm_runner(model,
+                     max_model_len=512,
+                     dtype=dtype,
+                     model_impl="transformers") as vllm_model:
+        vllm_outputs = vllm_model.classify(example_prompts)
+
+    with hf_runner(model,
+                   dtype=dtype,
+                   auto_cls=AutoModelForSequenceClassification) as hf_model:
+        hf_outputs = hf_model.classify(example_prompts)
+
+    for hf_output, vllm_output in zip(hf_outputs, vllm_outputs):
+        hf_output = torch.tensor(hf_output)
+        vllm_output = torch.tensor(vllm_output)
+
+        assert torch.allclose(hf_output, vllm_output,
+                              1e-3 if dtype == "float" else 1e-2)