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 8, 11:15 UTC
[GPT-OSS] Remove unnecessary contiguous (#34337)
Signed-off-by: elvischenv <219235043+elvischenv@users.noreply.github.com>
elvischenv committed
Feb 12, 2026 at 04:29 UTC
83e26c834ef188ca84b2459199840e2d58c75c32
1 file changed
-1
vllm/model_executor/models/gpt_oss.py
-1
@@ -140,7 +140,6 @@ class OAIAttention(nn.Module):
140
qkv, _ = self.qkv_proj(hidden_states)
141
q, k, v = qkv.split([self.q_size, self.kv_size, self.kv_size], dim=-1)
142
q, k = self.rotary_emb(positions, q, k)
143
- v = v.contiguous()
143
attn_output = self.attn(q, k, v)
144
output, _ = self.o_proj(attn_output)
145
return output