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, 21:30 UTC
[Bugfix] Fix MRotaryEmbedding missing `truncate` attr with YaRN scaling (#35080)
Signed-off-by: haosdent <haosdent@gmail.com>
haosdent committed
Feb 24, 2026 at 00:05 UTC
a7f341c32326fe5463556c69cfa54d90281041c8
1 file changed
+2
vllm/model_executor/layers/rotary_embedding/mrope.py
+2
@@ -218,12 +218,14 @@ class MRotaryEmbedding(RotaryEmbeddingBase):
218
attn_factor: float = 1,
219
beta_fast: int = 32,
220
beta_slow: int = 1,
221
+ truncate: bool = True,
222
) -> None:
223
self.scaling_factor = scaling_factor
224
self.extrapolation_factor = extrapolation_factor
225
self.attn_factor = attn_factor
226
self.beta_fast = beta_fast
227
self.beta_slow = beta_slow
228
+ self.truncate = truncate
229
if self.scaling_factor is not None:
230
# Get n-d magnitude scaling corrected for interpolation
231
self.mscale = float(yarn_get_mscale(self.scaling_factor) * attn_factor)