Update app.py

app.py

@@ -7,8 +7,10 @@ GB_S = 1935e9
 # in ms
 THREAD_OVERHEAD = 0.005
 
+# in ms
 def calc_exec_time(comp_flop, mem_bytes):
-  return (comp_flop/TFLOPS + mem_bytes/GB_S)*1000
+  exec_time = comp_flop/TFLOPS + mem_bytes/GB_S
+  return max(exec_time*1000, THREAD_OVERHEAD)
 
 def qkv_mha_exec(bs, h, n, d):
   flop = 2*bs*1*d*3*d
@@ -100,7 +102,7 @@ n = st.sidebar.number_input('End seq', value=1024)
 st.sidebar.header("GPU parameters")
 
 
-st.header("Total execution time")
+st.header("Execution time (ms)")
 
 mqa_total_time = 0.
 mha_total_time = 0.
@@ -113,8 +115,8 @@ for i in range(n_start, n):
   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]
   mqa_total_time += l*mqa_time
   
-st.write("MHA exec time: " + str(mha_total_time))
-st.write("MQA exec time: " + str(mqa_total_time))
+st.write("Multi-Head Attention: " + str(mha_total_time))
+st.write("Multi-Query Attention: " + str(mqa_total_time))
 
 st.header('Attention layer')