[Misc] Use VLLMValidationError in offline inference input validation (#53432)

Signed-off-by: Jensen Chen <a1043904820@gmail.com>
Co-authored-by: Claude Opus 5 <noreply@anthropic.com>

tests/entrypoints/llm/test_generate.py

@@ -6,6 +6,7 @@ import weakref
 import pytest
 
 from vllm import LLM, SamplingParams
+from vllm.exceptions import VLLMValidationError
 
 MODEL_NAME = "distilbert/distilgpt2"
 
@@ -52,7 +53,7 @@ def test_multiple_sampling_params(llm: LLM):
     assert len(PROMPTS) == len(outputs)
 
     # Exception raised, if the size of params does not match the size of prompts
-    with pytest.raises(ValueError):
+    with pytest.raises(VLLMValidationError):
         outputs = llm.generate(PROMPTS, sampling_params=sampling_params[:3])
 
     # Single SamplingParams should be applied to every prompt
@@ -75,13 +76,13 @@ def test_multiple_priority(llm: LLM):
     assert len(PROMPTS) == len(outputs)
 
     # Exception raised, if the length of priority does not match the length of prompts
-    with pytest.raises(ValueError):
+    with pytest.raises(VLLMValidationError):
         outputs = llm.generate(
             PROMPTS, sampling_params=None, priority=[0] * (len(PROMPTS) - 1)
         )
 
     # Exception raised, if the priority list is empty
-    with pytest.raises(ValueError):
+    with pytest.raises(VLLMValidationError):
         outputs = llm.generate(PROMPTS, sampling_params=None, priority=[])
 
 

tests/entrypoints/test_offline_utils.py

@@ -0,0 +1,57 @@
+# SPDX-License-Identifier: Apache-2.0
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
+
+import pytest
+
+from vllm import SamplingParams
+from vllm.entrypoints.offline_utils import OfflineInferenceMixin
+from vllm.exceptions import VLLMValidationError
+from vllm.lora.request import LoRARequest
+
+
+@pytest.fixture
+def mixin() -> OfflineInferenceMixin:
+    return object.__new__(OfflineInferenceMixin)
+
+
+def test_rejects_mismatched_params(mixin: OfflineInferenceMixin):
+    with pytest.raises(VLLMValidationError) as exc_info:
+        mixin._params_to_seq([SamplingParams()], num_requests=2)
+
+    assert str(exc_info.value) == (
+        "The lengths of prompts (2) and params (1) must be the same."
+    )
+    assert exc_info.value.parameter is None
+    assert exc_info.value.value is None
+
+
+def test_rejects_mismatched_lora_requests(mixin: OfflineInferenceMixin):
+    with pytest.raises(VLLMValidationError) as exc_info:
+        mixin._lora_request_to_seq([None], num_requests=2)
+
+    assert str(exc_info.value) == (
+        "The lengths of prompts (2) and lora_request (1) must be the same."
+    )
+    assert exc_info.value.parameter is None
+    assert exc_info.value.value is None
+
+
+def test_rejects_mismatched_priority(mixin: OfflineInferenceMixin):
+    with pytest.raises(VLLMValidationError) as exc_info:
+        mixin._priority_to_seq([0], num_requests=2)
+
+    assert str(exc_info.value) == (
+        "The lengths of prompts (2) and priority (1) must be the same."
+    )
+    assert exc_info.value.parameter is None
+    assert exc_info.value.value is None
+
+
+def test_matching_lengths_pass_through(mixin: OfflineInferenceMixin):
+    lora = LoRARequest("l", 1, "/tmp/l")
+
+    assert mixin._params_to_seq([SamplingParams()], num_requests=1) == [
+        SamplingParams()
+    ]
+    assert mixin._lora_request_to_seq([lora], num_requests=1) == [lora]
+    assert mixin._priority_to_seq([3], num_requests=1) == [3]

tests/lora/test_qwen3_with_multi_loras.py

@@ -12,6 +12,7 @@ import pytest
 
 from tests.utils import multi_gpu_test
 from vllm import LLM, SamplingParams
+from vllm.exceptions import VLLMValidationError
 from vllm.lora.request import LoRARequest
 
 MODEL_PATH = "Qwen/Qwen3-0.6B"
@@ -192,7 +193,7 @@ def test_multiple_lora_requests():
     assert len(PROMPTS) == len(outputs)
 
     # Exception raised, if the size of params does not match the size of prompts
-    with pytest.raises(ValueError):
+    with pytest.raises(VLLMValidationError):
         outputs = llm.generate(PROMPTS, lora_request=lora_request[:1])
 
     # Single LoRARequest should be applied to every prompt

vllm/entrypoints/offline_utils.py

@@ -19,6 +19,7 @@ from vllm.entrypoints.chat_utils import (
     ChatCompletionMessageParam,
     ChatTemplateContentFormatOption,
 )
+from vllm.exceptions import VLLMValidationError
 from vllm.inputs import EngineInput
 from vllm.logger import init_logger
 from vllm.lora.request import LoRARequest
@@ -246,7 +247,7 @@ class OfflineInferenceMixin:
     ) -> Sequence[_P]:
         if isinstance(params, Sequence):
             if len(params) != num_requests:
-                raise ValueError(
+                raise VLLMValidationError(
                     f"The lengths of prompts ({num_requests}) "
                     f"and params ({len(params)}) must be the same."
                 )
@@ -262,7 +263,7 @@ class OfflineInferenceMixin:
     ) -> Sequence[LoRARequest | None]:
         if isinstance(lora_request, Sequence):
             if len(lora_request) != num_requests:
-                raise ValueError(
+                raise VLLMValidationError(
                     f"The lengths of prompts ({num_requests}) "
                     f"and lora_request ({len(lora_request)}) must be the same."
                 )
@@ -278,7 +279,7 @@ class OfflineInferenceMixin:
     ) -> Sequence[int]:
         if priority is not None:
             if len(priority) != num_requests:
-                raise ValueError(
+                raise VLLMValidationError(
                     f"The lengths of prompts ({num_requests}) "
                     f"and priority ({len(priority)}) must be the same."
                 )