leslie-fang-intel

# Run with: clear && mpiexec -n 2 python test_mpi4.py
from mpi4py.futures import MPIPoolExecutor

def work(x):
    print(f"Working on: {x}")
    return x * x

if __name__ == "__main__":
    with MPIPoolExecutor() as executor:
        future = executor.submit(work, 5)

leslie-fang-intel

import time
import torch
from torch import nn

class M(torch.nn.Module):
    def __init__(self,):
        super().__init__()
        self.linear = torch.nn.Linear(1024, 512)
        self.relu = torch.nn.ReLU()

leslie-fang-intel

from setuptools import setup, find_packages, Extension, Command import glob import os import pybind11 import shutil

import torch from torch.utils.cpp_extension import ( CUDA_HOME, IS_WINDOWS,

leslie-fang-intel

import torch
import sglang
import sgl_kernel

if __name__ == "__main__":
    a = torch.randn((1, 1024), dtype=torch.float32).to("xpu")

    ref_res = a + a
    res3 = torch.ops.sgl_kernel.sgl_test_sycl(a, a)

leslie-fang-intel

import torch
import torch._inductor.config as config
# config.realize_opcount_threshold = 1

class SimpleModel(torch.nn.Module):
    def forward(self, x0, x1, x2):
        tmp = x0 + x1
        tmp2 = tmp * x2
        return tmp2

leslie-fang-intel

# TORCHINDUCTOR_FREEZING=1 TORCH_LOGS="+output_code" numactl -C 56-111 -m 1 python test.py

import torch
import time
import random
import numpy as np

local_seed= 2024

torch.manual_seed(local_seed) # Set PyTorch seed

leslie-fang-intel

import requests
import torch
print(torch.__version__, flush=True)
import torch.nn as nn
import os, pickle
import numpy as np 
import torch._inductor.config as config
import torch._dynamo.config as dynamo_config
import gc
import time

leslie-fang-intel

import requests
import torch
print(torch.__version__, flush=True)
import torch.nn as nn
import os, pickle
import numpy as np 
import torch._inductor.config as config
import torch._dynamo.config as dynamo_config
import gc
import time

leslie-fang-intel

import torch
from transformers import TorchAoConfig, AutoModelForCausalLM, AutoTokenizer
from torchao.quantization import int4_weight_only
from torchao.dtypes import Int4CPULayout
import torch._inductor.config as config

config.freezing = True
# config.max_autotune = True

with torch.no_grad():

leslie-fang-intel

import requests
import torch
print(torch.__version__, flush=True)
import torch.nn as nn
import os, pickle
import numpy as np 
import torch._inductor.config as config
import torch._dynamo.config as dynamo_config
import gc
import time