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 7, 22:45 UTC
[CLEANING] Remove unused disable_by_batch_size from SpeculativeConfig (#35060)
Signed-off-by: Vincent Gimenes <vincent.gimenes@gmail.com>
Vincent Gimenes committed
Feb 23, 2026 at 14:05 UTC
aa08a30fc90248006ce6202496926f074149b08c
1 file changed
-10
vllm/config/speculative.py
-10
@@ -101,9 +101,6 @@ class SpeculativeConfig:
101
will use the default version."""
102
103
# Advanced control
104
- disable_by_batch_size: int | None = Field(default=None, ge=2)
105
- """Disable speculative decoding for new incoming requests when the number
106
- of enqueued requests is larger than this value, if provided."""
104
disable_padded_drafter_batch: bool = False
105
"""Disable input padding for speculative decoding. If set to True,
106
speculative input batches can contain sequences of different lengths,
@@ -707,13 +704,6 @@ class SpeculativeConfig:
704
self.draft_parallel_config
705
)
706
710
- if self.disable_by_batch_size is not None and self.disable_by_batch_size < 2:
711
- raise ValueError(
712
- "Expect the batch size threshold of disabling "
713
- "speculative decoding is > 1, but got "
714
- f"{self.disable_by_batch_size=}"
715
- )
716
-
707
eagle3_target_supported = [
708
"llama",
709
"qwen",