Update app.py

app.py

@@ -108,8 +108,8 @@ mqa_total_time = 0.
 mha_total_time = 0.
 
 for i in range(n_start, n):
-  shared_time = out_exec(bs, h, i, d)[2] + softmax_exec(bs, h, i , d)[2] \ 
-                2*ln_exec(bs, h, i, d)[2] + 2*mlp_exec(bs, h, i, d)[2] + 3*ln_exec(bs, h, i, d)
+  shared_time = out_exec(bs, h, i, d)[2] + softmax_exec(bs, h, i , d)[2] + 2*ln_exec(bs, h, i, d)[2] \ 
+                + 2*mlp_exec(bs, h, i, d)[2] + 3*ln_exec(bs, h, i, d)
   mha_time = shared_time + qkv_mha_exec(bs, h, i, d)[2] + att1_mha_exec(bs, h, i, d)[2] + att2_mha_exec(bs, h, i, d)[2]
   mha_total_time += l*mha_time
   mqa_time = shared_time + qkv_mqa_exec(bs, h, i, d)[2] + att1_mqa_exec(bs, h, i, d)[2] + att2_mqa_exec(bs, h, i, d)[2]