performance of `position_ids`

position_ids_perf.py

#%%
import paddle
from paddle import nn
import time
import pandas as pd
from tqdm import tqdm, trange
                



def perf_position_ids_cache(epoch: int = 1000, max_position_embedding: int = 512, hidden_size: int = 768):
    position_ids_cache = paddle.expand(paddle.arange(max_position_embedding, dtype="int64"),
                            shape=[1, -1])

    start = time.perf_counter()
    for _ in range(epoch):
        _ = position_ids_cache[: max_position_embedding // 2]
    cost = (time.perf_counter() - start) / 50 * 1000
    return cost

def perf_position_ids(epoch: int = 1000, max_position_embedding: int = 512, hidden_size: int = 768):
    seq_len = max_position_embedding // 2
    input_ids = paddle.randn([4, seq_len])

    start = time.perf_counter()
    for _ in range(epoch):
        ones = paddle.ones_like(input_ids, dtype="int64")
        seq_length = paddle.cumsum(ones, axis=-1)
        position_ids = seq_length - ones 
    cost = (time.perf_counter() - start) / 50 * 1000
    return cost

#%%
def run_function_perf():
    # epoch
    series = []
    for i in trange(100, 1000, 100):
        cache_cost = perf_position_ids_cache(epoch=i)
        no_cache_cost = perf_position_ids(epoch=i)

        series.append(
            {
                "cache": cache_cost,
                "no_cache": no_cache_cost,
                "speed up": (no_cache_cost / cache_cost),
                "epochs": i
            }
        )
    table = pd.DataFrame(series)
    # table.plot(x="epochs", y=['cache', "no_cache"])
    table.plot(x="epochs", y="speed up")
    # print(table)
    
    # # epoch
    # series = []
    # for i in trange(128, 1024, 10):
    #     series.append(
    #         {
    #             "cache": perf_position_ids_cache(max_position_embedding=i),
    #             "no_cache": perf_position_ids(max_position_embedding=i)
    #         }
    #     )
    # pd.DataFrame(series).to_excel("./sss/perf-max-posi.xlsx", index=False)
        
run_function_perf()