[Perf] Reduce peak memory usage of llama (#10339)

Signed-off-by: andoorve <37849411+andoorve@users.noreply.github.com>

vllm/model_executor/models/llama.py

@@ -90,8 +90,8 @@ class LlamaMLP(nn.Module):
         self.act_fn = SiluAndMul()
 
     def forward(self, x):
-        gate_up, _ = self.gate_up_proj(x)
-        x = self.act_fn(gate_up)
+        x, _ = self.gate_up_proj(x)
+        x = self.act_fn(x)
         x, _ = self.down_proj(x)
         return x