Skip to content

Commit 46b4a02

Browse files
catswelalo
andauthored
Fix DeepSeek V2 RoPE initialization error (vllm-project#33501)
Signed-off-by: Eduardo Salinas <[email protected]> Signed-off-by: catswe <[email protected]> Co-authored-by: Eduardo Salinas <[email protected]>
1 parent 8869cd8 commit 46b4a02

File tree

1 file changed

+0
-2
lines changed

1 file changed

+0
-2
lines changed

vllm/model_executor/layers/rotary_embedding/deepseek_scaling_rope.py

Lines changed: 0 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -75,7 +75,6 @@ def _compute_inv_freq(self, scaling_factor: float) -> torch.Tensor:
7575
self.rotary_dim,
7676
2,
7777
dtype=torch.float,
78-
device=current_platform.device_type,
7978
)
8079
/ self.rotary_dim
8180
)
@@ -104,7 +103,6 @@ def _compute_cos_sin_cache(self) -> torch.Tensor:
104103
inv_freq = self._compute_inv_freq(self.scaling_factor)
105104
t = torch.arange(
106105
self.max_position_embeddings * self.scaling_factor,
107-
device=current_platform.device_type,
108106
dtype=torch.float32,
109107
)
110108
freqs = torch.einsum("i,j -> ij", t, inv_freq)

0 commit comments

Comments
 (0)