Spaces:
Build error
Build error
File size: 5,300 Bytes
546a9ba |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 |
# Copyright (c) Microsoft Corporation.
# Licensed under the MIT license.
import os
import torch
from torch.utils.tensorboard import SummaryWriter
import random
import numpy as np
from pkg_resources import parse_version
from model.third_party.HMNet.Models.Trainers.BaseTrainer import BaseTrainer
from model.third_party.HMNet.Utils.GeneralUtils import bcolors
from model.third_party.HMNet.Utils.distributed import distributed
class DistributedTrainer(BaseTrainer):
def __init__(self, opt):
super().__init__(opt)
self.seed = int(self.opt["SEED"]) if "SEED" in self.opt else 0
random.seed(self.seed)
np.random.seed(self.seed)
torch.manual_seed(self.seed)
(
self.opt["device"],
_,
self.opt["world_size"],
self.opt["local_size"],
self.opt["rank"],
self.opt["local_rank"],
_,
self.opt["run"],
) = distributed(opt, not self.use_cuda)
self.getSaveFolder()
self.opt["logFile"] = f"log_{self.opt['rank']}.txt"
self.saveConf()
self.high_pytorch_version = parse_version(torch.__version__) >= parse_version(
"1.2.0"
)
if self.opt["rank"] == 0:
print(
bcolors.OKGREEN,
torch.__version__,
bcolors.ENDC,
"is",
"high" if self.high_pytorch_version else "low",
)
if self.use_cuda:
# torch.cuda.manual_seed_all(self.seed)
# ddp: only set seed on GPU associated with this process
torch.cuda.manual_seed(self.seed)
# ddp: print stats and update learning rate
if self.opt["rank"] == 0:
print(
"Number of GPUs is",
bcolors.OKGREEN,
self.opt["world_size"],
bcolors.ENDC,
)
# print('Boost learning rate from', bcolors.OKGREEN, self.opt['START_LEARNING_RATE'], bcolors.ENDC, 'to',
# bcolors.OKGREEN, self.opt['START_LEARNING_RATE'] * self.opt['world_size'], bcolors.ENDC)
print(
"Effective batch size is increased from",
bcolors.OKGREEN,
self.opt["MINI_BATCH"],
bcolors.ENDC,
"to",
bcolors.OKGREEN,
self.opt["MINI_BATCH"] * self.opt["world_size"],
bcolors.ENDC,
)
self.grad_acc_steps = 1
if "GRADIENT_ACCUMULATE_STEP" in self.opt:
if self.opt["rank"] == 0:
print(
"Gradient accumulation steps =",
bcolors.OKGREEN,
self.opt["GRADIENT_ACCUMULATE_STEP"],
bcolors.ENDC,
)
# print('Boost learning rate from', bcolors.OKGREEN, self.opt['START_LEARNING_RATE'], bcolors.ENDC, 'to',
# bcolors.OKGREEN, self.opt['START_LEARNING_RATE'] * self.opt['world_size'] * self.opt['GRADIENT_ACCUMULATE_STEP'], bcolors.ENDC)
print(
"Effective batch size =",
bcolors.OKGREEN,
self.opt["MINI_BATCH"]
* self.opt["world_size"]
* self.opt["GRADIENT_ACCUMULATE_STEP"],
bcolors.ENDC,
)
self.grad_acc_steps = int(self.opt["GRADIENT_ACCUMULATE_STEP"])
# self.opt['START_LEARNING_RATE'] *= self.opt['world_size'] * self.grad_acc_steps
def tb_log_scalar(self, name, value, step):
if self.opt["rank"] == 0:
if self.tb_writer is None:
self.tb_writer = SummaryWriter(
os.path.join(self.saveFolder, "tensorboard")
)
self.tb_writer.add_scalar(name, value, step)
def log(self, s):
# When 'OFFICIAL' flag is set in the config file, the program does not output logs
if self.is_official:
return
try:
if self.logFileHandle is None:
self.logFileHandle = open(
os.path.join(self.saveFolder, self.opt["logFile"]), "a"
)
self.logFileHandle.write(s + "\n")
except Exception as e:
print("ERROR while writing log file:", e)
print(s)
def getSaveFolder(self):
runid = 1
while True:
saveFolder = os.path.join(
self.opt["datadir"],
self.opt["basename"] + "_conf~",
"run_" + str(runid),
)
if not os.path.isdir(saveFolder):
if self.opt["world_size"] > 1:
torch.distributed.barrier()
if self.opt["rank"] == 0:
os.makedirs(saveFolder)
self.saveFolder = saveFolder
if self.opt["world_size"] > 1:
torch.distributed.barrier()
print(
"Saving logs, model, checkpoint, and evaluation in "
+ self.saveFolder
)
return
runid = runid + 1
def saveConf(self):
if self.opt["rank"] == 0:
super().saveConf()
|