From 2f5c14de6a7e281a86ab5e2376de95f7021dff0a Mon Sep 17 00:00:00 2001
From: youkaichao <youkaichao@gmail.com>
Date: Wed, 23 Jul 2025 15:03:16 +0800
Subject: [PATCH] add clear messages for deprecated models (#21424)

Signed-off-by: youkaichao <youkaichao@gmail.com>
---
 vllm/model_executor/model_loader/utils.py | 11 ++++++++++-
 vllm/model_executor/models/registry.py    |  2 ++
 2 files changed, 12 insertions(+), 1 deletion(-)

diff --git a/vllm/model_executor/model_loader/utils.py b/vllm/model_executor/model_loader/utils.py
index 42c5512905f2..4b30336f0132 100644
--- a/vllm/model_executor/model_loader/utils.py
+++ b/vllm/model_executor/model_loader/utils.py
@@ -25,7 +25,8 @@ from vllm.model_executor.models.adapters import (as_embedding_model,
                                                  as_reward_model,
                                                  as_seq_cls_model)
 from vllm.model_executor.models.interfaces import SupportsQuant
-from vllm.model_executor.models.registry import _TRANSFORMERS_MODELS
+from vllm.model_executor.models.registry import (_PREVIOUSLY_SUPPORTED_MODELS,
+                                                 _TRANSFORMERS_MODELS)
 from vllm.utils import is_pin_memory_available
 
 logger = init_logger(__name__)
@@ -261,6 +262,14 @@ def get_model_architecture(
             vllm_not_supported = False
             break
 
+    if any(arch in _PREVIOUSLY_SUPPORTED_MODELS for arch in architectures):
+        previous_version = _PREVIOUSLY_SUPPORTED_MODELS[architectures[0]]
+        raise ValueError(
+            f"Model architecture {architectures[0]} was supported"
+            f" in vLLM until version {previous_version}, and is "
+            "not supported anymore. Please use an older version"
+            " of vLLM if you want to use this model architecture.")
+
     if (model_config.model_impl == ModelImpl.TRANSFORMERS or
             model_config.model_impl == ModelImpl.AUTO and vllm_not_supported):
         architectures = resolve_transformers_arch(model_config, architectures)
diff --git a/vllm/model_executor/models/registry.py b/vllm/model_executor/models/registry.py
index 9d88b5fe82cf..100532943c2b 100644
--- a/vllm/model_executor/models/registry.py
+++ b/vllm/model_executor/models/registry.py
@@ -276,6 +276,8 @@ _SUBPROCESS_COMMAND = [
     sys.executable, "-m", "vllm.model_executor.models.registry"
 ]
 
+_PREVIOUSLY_SUPPORTED_MODELS = {"Phi3SmallForCausalLM": "0.9.2"}
+
 
 @dataclass(frozen=True)
 class _ModelInfo: