fix(security): Add VLLM_MAX_N_SEQUENCES environment variable and enforce limit (#37952)

Signed-off-by: jperezde <jperezde@redhat.com>
Signed-off-by: Russell Bryant <rbryant@redhat.com>
Co-authored-by: Russell Bryant <rbryant@redhat.com>
This commit is contained in:
Juan Pérez de Algaba
2026-03-27 09:02:10 -04:00
committed by GitHub
co-authored by Russell Bryant
parent 497e234d38
commit b111f8a61f
5 changed files with 193 additions and 0 deletions
+11
View File
@@ -12,6 +12,7 @@ from typing import Any
import msgspec
from pydantic.dataclasses import dataclass
import vllm.envs as envs
from vllm.config import ModelConfig, SpeculativeConfig, StructuredOutputsConfig
from vllm.exceptions import VLLMValidationError
from vllm.logger import init_logger
@@ -169,6 +170,9 @@ class SamplingParams(
n: int = 1
"""Number of outputs to return for the given prompt request.
The maximum allowed value is controlled by the ``VLLM_MAX_N_SEQUENCES``
environment variable (default: 16384).
NOTE:
`AsyncLLM` streams outputs by default. When `n > 1`, all `n` outputs
are generated and streamed cumulatively per request. To see all `n`
@@ -425,6 +429,13 @@ class SamplingParams(
raise ValueError(f"n must be an int, but is of type {type(self.n)}")
if self.n < 1:
raise ValueError(f"n must be at least 1, got {self.n}.")
max_n = envs.VLLM_MAX_N_SEQUENCES
if self.n > max_n:
raise ValueError(
f"n must be at most {max_n}, got {self.n}. "
"To increase this limit, set the VLLM_MAX_N_SEQUENCES "
"environment variable."
)
if not -2.0 <= self.presence_penalty <= 2.0:
raise ValueError(
f"presence_penalty must be in [-2, 2], got {self.presence_penalty}."