diff --git a/vllm/model_executor/layers/rotary_embedding/mrope.py b/vllm/model_executor/layers/rotary_embedding/mrope.py index 52f3c333d..3c946dd13 100644 --- a/vllm/model_executor/layers/rotary_embedding/mrope.py +++ b/vllm/model_executor/layers/rotary_embedding/mrope.py @@ -218,12 +218,14 @@ class MRotaryEmbedding(RotaryEmbeddingBase): attn_factor: float = 1, beta_fast: int = 32, beta_slow: int = 1, + truncate: bool = True, ) -> None: self.scaling_factor = scaling_factor self.extrapolation_factor = extrapolation_factor self.attn_factor = attn_factor self.beta_fast = beta_fast self.beta_slow = beta_slow + self.truncate = truncate if self.scaling_factor is not None: # Get n-d magnitude scaling corrected for interpolation self.mscale = float(yarn_get_mscale(self.scaling_factor) * attn_factor)