A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 11, 03:45 UTC
[Bugfix] Read truncate_prompt_tokens from pooling_params in AsyncLLM.encode() (#31013)
Signed-off-by: Jeffrey Wang <jeffreywang@anyscale.com>
Jeffrey Wang committed
Dec 20, 2025 at 02:29 UTC
1501a4070ec1f5e3f8f1ffd0099dc32d85a9ad98
2 files changed
+20
-2
vllm/engine/protocol.py
+5
-1
@@ -71,7 +71,11 @@ class EngineClient(ABC):
71
truncate_prompt_tokens: int | None = None,
72
tokenization_kwargs: dict[str, Any] | None = None,
73
) -> AsyncGenerator[PoolingRequestOutput, None]:
74
- """Generate outputs for a request from a pooling model."""
74
+ """Generate outputs for a request from a pooling model.
75
+
76
+ NOTE: truncate_prompt_tokens is deprecated in v0.14.
77
+ TODO: Remove this argument in v0.15.
78
+ """
79
...
80
81
@abstractmethod
vllm/v1/engine/async_llm.py
+15
-1
@@ -4,6 +4,7 @@ import asyncio
4
import os
5
import socket
6
import time
7
+import warnings
8
from collections.abc import AsyncGenerator, Iterable, Mapping
9
from copy import copy
10
from typing import Any, cast
@@ -627,6 +628,9 @@ class AsyncLLM(EngineClient):
628
629
The caller of generate() iterates the returned AsyncGenerator,
630
returning the RequestOutput back to the caller.
631
+
632
+ NOTE: truncate_prompt_tokens is deprecated in v0.14.
633
+ TODO: Remove truncate_prompt_tokens in v0.15.
634
"""
635
636
try:
@@ -641,9 +645,19 @@ class AsyncLLM(EngineClient):
645
646
if tokenization_kwargs is None:
647
tokenization_kwargs = {}
648
+
649
+ if truncate_prompt_tokens is not None:
650
+ warnings.warn(
651
+ "The `truncate_prompt_tokens` parameter in `AsyncLLM.encode()` "
652
+ "is deprecated and will be removed in v0.15. "
653
+ "Please use `pooling_params.truncate_prompt_tokens` instead.",
654
+ DeprecationWarning,
655
+ stacklevel=2,
656
+ )
657
+
658
_validate_truncation_size(
659
self.model_config.max_model_len,
646
- truncate_prompt_tokens,
660
+ pooling_params.truncate_prompt_tokens,
661
tokenization_kwargs,
662
)
663