[Frontend][Core] Standardize request error handling with VLLMError hierarchy (#49665)

Signed-off-by: Zach Zhu <zzqshu@126.com>
This commit is contained in:
Zach Zhu
2026-07-29 04:20:40 +00:00
committed by GitHub
parent f37f03db4a
commit 58f9659397
46 changed files with 370 additions and 248 deletions
+2 -1
View File
@@ -6,6 +6,7 @@ import pytest
from vllm import LLM
from vllm.distributed import cleanup_dist_env_and_memory
from vllm.exceptions import VLLMValidationError
from vllm.sampling_params import SamplingParams
@@ -157,7 +158,7 @@ def test_chat_batch_failure_cleanup(llm_for_failure_test):
batch_2 = [valid_msg, valid_msg]
sampling_params = SamplingParams(temperature=0, max_tokens=10)
with pytest.raises(ValueError, match="maximum context length is"):
with pytest.raises(VLLMValidationError, match="maximum context length is"):
llm.chat(batch_1, sampling_params=sampling_params)
assert llm.llm_engine.get_num_unfinished_requests() == 0
@@ -5,17 +5,18 @@ import pytest
import torch
from vllm import LLM
from vllm.exceptions import VLLMValidationError
def test_empty_prompt():
llm = LLM(model="openai-community/gpt2", enforce_eager=True)
with pytest.raises(ValueError, match="decoder prompt cannot be empty"):
with pytest.raises(VLLMValidationError, match="decoder prompt cannot be empty"):
llm.generate([""])
def test_out_of_vocab_token():
llm = LLM(model="openai-community/gpt2", enforce_eager=True)
with pytest.raises(ValueError, match="out of vocabulary"):
with pytest.raises(VLLMValidationError, match="out of vocabulary"):
llm.generate({"prompt_token_ids": [999999]})