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 15, 08:30 UTC
[BUG FIX] minicpm (#18739)
Signed-off-by: huangyuxiang03 <huangyx0321@gmail.com> Co-authored-by: huangyuxiang03 <huangyx0321@gmail.com>
Shawn Huang committed
May 27, 2025 at 16:04 UTC
6881107948c00a8564bc2fa85308f6fc2f065d64
1 file changed
-3
vllm/model_executor/models/minicpm.py
-3
@@ -242,9 +242,6 @@ class MiniCPMAttention(nn.Module):
242
base=rope_theta,
243
rope_scaling=rope_scaling,
244
)
245
- # set rope as fp32 instead of bf16
246
- self.rotary_emb.cos_sin_cache = self.rotary_emb._compute_cos_sin_cache(
247
- )
245
self.attn = Attention(self.num_heads,
246
self.head_dim,
247
self.scaling,