按照 pip 方式安装后任何方式皆无法运行，下为执行日志，系统为 Arch Linux，Linux 系统内核版本为 Linux 5.13 rc4

>>> import jittor
[i 0604 13:52:10.823614 92 compiler.py:857] Jittor(1.2.3.14) src: /home/peter/.local/lib/python3.8/site-packages/jittor
[i 0604 13:52:10.828678 92 compiler.py:858] g++ at /usr/bin/g++(11.1.0)
[i 0604 13:52:10.828826 92 compiler.py:859] cache_path: /home/peter/.cache/jittor/default/g++
[i 0604 13:52:10.839448 92 __init__.py:258] Found nvcc(11.3.58) at /opt/cuda/bin/nvcc.
[i 0604 13:52:10.963726 92 __init__.py:258] Found gdb(10.2) at /usr/bin/gdb.
[i 0604 13:52:10.970791 92 __init__.py:258] Found addr2line(2.36.1) at /usr/bin/addr2line.
[i 0604 13:52:10.999637 92 compiler.py:915] py_include: -I/opt/anaconda/include/python3.8 -I/opt/anaconda/include/python3.8
[i 0604 13:52:11.027781 92 compiler.py:917] extension_suffix: .cpython-38-x86_64-linux-gnu.so
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/home/peter/.local/lib/python3.8/site-packages/jittor/__init__.py", line 18, in <module>
    from . import compiler
  File "/home/peter/.local/lib/python3.8/site-packages/jittor/compiler.py", line 929, in <module>
    check_cache_compile()
  File "/home/peter/.local/lib/python3.8/site-packages/jittor/compiler.py", line 800, in check_cache_compile
    assert jit_utils.cc
AssertionError

I followed the install instruction to install llvm on Ubuntu 16.04 but failed with

Ign:27 http://ppa.launchpad.net/mc3man/trusty-media/ubuntu xenial/main i386 Packages
Ign:28 http://ppa.launchpad.net/mc3man/trusty-media/ubuntu xenial/main all Packages
Ign:29 http://ppa.launchpad.net/mc3man/trusty-media/ubuntu xenial/main Translation-en_US
Ign:30 http://ppa.launchpad.net/mc3man/trusty-media/ubuntu xenial/main Translation-en
Ign:31 http://ppa.launchpad.net/mc3man/trusty-media/ubuntu xenial/main amd64 DEP-11 Metadata
Ign:32 http://ppa.launchpad.net/mc3man/trusty-media/ubuntu xenial/main DEP-11 64x64 Icons
Ign:33 http://ppa.launchpad.net/jonathonf/python-3.6/ubuntu xenial/main amd64 Packages
Ign:34 http://ppa.launchpad.net/jonathonf/python-3.6/ubuntu xenial/main i386 Packages
Ign:21 http://ppa.launchpad.net/jonathonf/python-3.6/ubuntu xenial/main all Packages
Ign:22 http://ppa.launchpad.net/jonathonf/python-3.6/ubuntu xenial/main Translation-en_US
Ign:35 http://ppa.launchpad.net/jonathonf/python-3.6/ubuntu xenial/main Translation-en
Ign:24 http://ppa.launchpad.net/jonathonf/python-3.6/ubuntu xenial/main amd64 DEP-11 Metadata
Ign:25 http://ppa.launchpad.net/jonathonf/python-3.6/ubuntu xenial/main DEP-11 64x64 Icons
Err:26 http://ppa.launchpad.net/mc3man/trusty-media/ubuntu xenial/main amd64 Packages
  404  Not Found [IP: 2001:67c:1560:8008::15 80]
Ign:27 http://ppa.launchpad.net/mc3man/trusty-media/ubuntu xenial/main i386 Packages
Ign:28 http://ppa.launchpad.net/mc3man/trusty-media/ubuntu xenial/main all Packages
Ign:29 http://ppa.launchpad.net/mc3man/trusty-media/ubuntu xenial/main Translation-en_US
Ign:30 http://ppa.launchpad.net/mc3man/trusty-media/ubuntu xenial/main Translation-en
Ign:31 http://ppa.launchpad.net/mc3man/trusty-media/ubuntu xenial/main amd64 DEP-11 Metadata
Ign:32 http://ppa.launchpad.net/mc3man/trusty-media/ubuntu xenial/main DEP-11 64x64 Icons
Err:33 http://ppa.launchpad.net/jonathonf/python-3.6/ubuntu xenial/main amd64 Packages
  403  Forbidden [IP: 2001:67c:1560:8008::15 80]
Ign:34 http://ppa.launchpad.net/jonathonf/python-3.6/ubuntu xenial/main i386 Packages
Ign:35 http://ppa.launchpad.net/jonathonf/python-3.6/ubuntu xenial/main Translation-en
Reading package lists... Done
W: The repository 'http://ppa.launchpad.net/jonathonf/python-3.6/ubuntu xenial Release' does not have a Release file.
N: Data from such a repository can't be authenticated and is therefore potentially dangerous to use.
N: See apt-secure(8) manpage for repository creation and user configuration details.
W: The repository 'http://ppa.launchpad.net/mc3man/trusty-media/ubuntu xenial Release' does not have a Release file.
N: Data from such a repository can't be authenticated and is therefore potentially dangerous to use.
N: See apt-secure(8) manpage for repository creation and user configuration details.
E: Failed to fetch http://ppa.launchpad.net/mc3man/trusty-media/ubuntu/dists/xenial/main/binary-amd64/Packages  404  Not Found [IP: 2001:67c:1560:8008::15 80]
E: Failed to fetch http://ppa.launchpad.net/jonathonf/python-3.6/ubuntu/dists/xenial/main/binary-amd64/Packages  403  Forbidden [IP: 2001:67c:1560:8008::15 80]
E: Some index files failed to download. They have been ignored, or old ones used instead.

How to solve this problem?

执行顺序 git clone https://github.com/Jittor/jittor.git sudo pip install ./jittor

python -m jittor.test.test_example

错误 urllib.error.URLError: <urlopen error [SSL: CERTIFICATE_VERIFY_FAILED] certificate verify failed: unable to get local issuer certificate (_ssl.c:1108)>

注释掉以下代码后 version_file = os.path.join(jittor_path, "version") if os.path.isfile(version_file) and not os.path.isdir(os.path.join(jittor_path, "src", "data")): with open(version_file, 'r') as f: version = f.read().strip() # key = f"{version}-{cc_type}-{'cuda' if has_cuda else 'cpu'}.o" key = f"{version}-g++-cpu" os_id = os_release["ID"] os_key = os_type.get(os_id, "ubuntu") if "os_key" in os.environ: os_key = os.environ['os_key'] if platform.machine()=='aarch64': os_key += '-aarch64' LOG.i("OS type:", os_id, " OS key:", os_key) key += '-' + os_key + '.o' # TODO: open the website extra_obj = os.path.join(cache_path, key) url = os.path.join("https://cg.cs.tsinghua.edu.cn/jittor/assets/build/"+key) jit_utils.download(url, extra_obj) files.append(extra_obj)

另一个错误 ImportError: .cache/jittor/default/g++/jittor_core.cpython-38-x86_64-linux-gnu.so: undefined symbol: _ZN6jittor4Node11tflag_countE

1.模型(ResNet)代码：

import jittor as jt
from jittor import nn
from jittor import Module
from jittor import init
from jittor.contrib import concat, argmax_pool
import time

__all__ = ["ResNet", "ResNet18", "ResNet34", "ResNet50", "ResNet101", "ResNet152"]


class BasicBlock(Module):
    expansion = 1

    def __init__(self, inplanes, planes, stride=1, downsample=None):
        self.conv1 = nn.Conv(inplanes, planes, kernel_size=3, stride=stride, padding=1, bias=False)
        self.bn1 = nn.BatchNorm(planes)
        self.relu = nn.Relu()
        self.conv2 = nn.Conv(planes, planes, kernel_size=3, stride=1, padding=1, bias=False)
        self.bn2 = nn.BatchNorm(planes)
        self.downsample = downsample
        self.stride = stride
        self.planes = planes

    def execute(self, x):
        residual = x
        out = self.conv1(x)
        out = self.bn1(out)
        out = self.relu(out)
        out = self.conv2(out)
        out = self.bn2(out)

        if self.downsample is not None:
            residual = self.downsample(x)

        out += residual
        out = self.relu(out)
        return out


class Bottleneck(Module):
    expansion = 4

    def __init__(self, inplanes, planes, stride=1, downsample=None):
        self.conv1 = nn.Conv(inplanes, planes, kernel_size=1, bias=False)
        self.bn1 = nn.BatchNorm(planes)
        self.conv2 = nn.Conv(planes, planes, kernel_size=3, stride=stride, padding=1, bias=False)
        self.bn2 = nn.BatchNorm(planes)
        self.conv3 = nn.Conv(planes, planes * self.expansion, kernel_size=1, bias=False)
        self.bn3 = nn.BatchNorm(planes * self.expansion)
        self.relu = nn.Relu()
        self.downsample = downsample
        self.stride = stride

    def execute(self, x):
        residual = x

        out = self.conv1(x)
        out = self.bn1(out)
        out = self.relu(out)

        out = self.conv2(out)
        out = self.bn2(out)
        out = self.relu(out)

        out = self.conv3(out)
        out = self.bn3(out)

        if self.downsample is not None:
            residual = self.downsample(x)

        out += residual
        out = self.relu(out)
        return out


class ResNet(Module):
    def __init__(self, block, layers, num_classes=1000):
        self.inplanes = 64
        self.conv1 = nn.Conv(3, 64, kernel_size=7, stride=2, padding=3, bias=False)
        self.bn1 = nn.BatchNorm(64)
        self.relu = nn.Relu()
        self.maxpool = nn.Pool(kernel_size=3, stride=2, padding=1)
        self.layer1 = self._make_layer(block, 64, layers[0])
        self.layer2 = self._make_layer(block, 128, layers[1], stride=2)
        self.layer3 = self._make_layer(block, 256, layers[2], stride=2)
        self.layer4 = self._make_layer(block, 512, layers[3], stride=2)
        self.avgpool = nn.Pool(7, stride=1, op="mean")
        self.fc = nn.Linear(512 * block.expansion, num_classes)

    def _make_layer(self, block, planes, blocks, stride=1):
        downsample = None
        if stride != 1 or self.inplanes != planes * block.expansion:
            downsample = nn.Sequential(
                nn.Conv(self.inplanes, planes * block.expansion,
                        kernel_size=1, stride=stride, bias=False),
                nn.BatchNorm(planes * block.expansion),
            )

        layers = []
        layers.append(block(self.inplanes, planes, stride, downsample))
        self.inplanes = planes * block.expansion
        for i in range(1, blocks):
            layers.append(block(self.inplanes, planes))

        return nn.Sequential(*layers)

    def execute(self, x):
        x = self.conv1(x)
        x = self.bn1(x)
        x = self.relu(x)
        x = self.maxpool(x)
        x = self.layer1(x)
        x = self.layer2(x)
        x = self.layer3(x)
        x = self.layer4(x)

        x = self.avgpool(x)
        x = jt.reshape(x, [x.shape[0], -1])
        x = self.fc(x)

        return x


def ResNet18():
    model = ResNet(BasicBlock, [2, 2, 2, 2])
    return model


def ResNet34():
    model = ResNet(BasicBlock, [3, 4, 6, 3])
    return model


def ResNet50():
    model = ResNet(Bottleneck, [3, 4, 6, 3])
    return model


def ResNet101():
    model = ResNet(Bottleneck, [3, 4, 23, 3])
    return model


def ResNet152():
    model = ResNet(Bottleneck, [3, 8, 36, 3])
    return model

2.模型VGG代码

# ***************************************************************
# Copyright (c) 2020 Jittor. Authors:
#     Guoye Yang <[email protected]>
#     Dun Liang <[email protected]>.
# All Rights Reserved.
# This file is subject to the terms and conditions defined in
# file 'LICENSE.txt', which is part of this source code package.
# ***************************************************************
import jittor as jt
from jittor import nn


__all__ = ['VGG', 'VGG11', 'VGG11_bn', 'VGG13', 'VGG13_bn', 'VGG16', 'VGG16_bn',
    'VGG19_bn', 'VGG19']


class VGG(nn.Module):

    def __init__(self, features, num_classes=1000, init_weights=True):
        super(VGG, self).__init__()
        self.features = features
        self.classifier = nn.Sequential(
            nn.Linear(512 * 7 * 7, 4096),
            nn.ReLU(),
            nn.Dropout(),
            nn.Linear(4096, 4096),
            nn.ReLU(),
            nn.Dropout(),
            nn.Linear(4096, num_classes),
        )

    def execute(self, x):
        x = self.features(x)
        x = jt.reshape(x, [x.shape[0],-1])
        x = self.classifier(x)
        return x

def make_layers(cfg, batch_norm=False):
    layers = []
    in_channels = 3
    for v in cfg:
        if v == 'M':
            layers += [nn.Pool(kernel_size=2, stride=2, op="maximum")]
        else:
            conv2d = nn.Conv(in_channels, v, kernel_size=3, padding=1)
            if batch_norm:
                layers += [conv2d, nn.BatchNorm(v), nn.ReLU()]
            else:
                layers += [conv2d, nn.ReLU()]
            in_channels = v
    return nn.Sequential(*layers)


cfgs = {
    'A': [64, 'M', 128, 'M', 256, 256, 'M', 512, 512, 'M', 512, 512, 'M'],
    'B': [64, 64, 'M', 128, 128, 'M', 256, 256, 'M', 512, 512, 'M', 512, 512, 'M'],
    'D': [64, 64, 'M', 128, 128, 'M', 256, 256, 256, 'M', 512, 512, 512, 'M', 512, 512, 512, 'M'],
    'E': [64, 64, 'M', 128, 128, 'M', 256, 256, 256, 256, 'M', 512, 512, 512, 512, 'M', 512, 512, 512, 512, 'M'],
}


def _vgg(arch, cfg, batch_norm, **kwargs):
    model = VGG(make_layers(cfgs[cfg], batch_norm=batch_norm), **kwargs)
    return model


def VGG11(**kwargs):
    r"""VGG 11-layer model (configuration "A") from
    `"Very Deep Convolutional Networks For Large-Scale Image Recognition" <https://arxiv.org/pdf/1409.1556.pdf>`_
    """
    return _vgg('vgg11', 'A', False, **kwargs)


def VGG11_bn(**kwargs):
    r"""VGG 11-layer model (configuration "A") with batch normalization
    `"Very Deep Convolutional Networks For Large-Scale Image Recognition" <https://arxiv.org/pdf/1409.1556.pdf>`_
    """
    return _vgg('vgg11_bn', 'A', True, **kwargs)


def VGG13(**kwargs):
    r"""VGG 13-layer model (configuration "B")
    `"Very Deep Convolutional Networks For Large-Scale Image Recognition" <https://arxiv.org/pdf/1409.1556.pdf>`_
    """
    return _vgg('vgg13', 'B', False, **kwargs)


def VGG13_bn(**kwargs):
    r"""VGG 13-layer model (configuration "B") with batch normalization
    `"Very Deep Convolutional Networks For Large-Scale Image Recognition" <https://arxiv.org/pdf/1409.1556.pdf>`_
    """
    return _vgg('vgg13_bn', 'B', True, **kwargs)


def VGG16(**kwargs):
    r"""VGG 16-layer model (configuration "D")
    `"Very Deep Convolutional Networks For Large-Scale Image Recognition" <https://arxiv.org/pdf/1409.1556.pdf>`_
    """
    return _vgg('vgg16', 'D', False, **kwargs)


def VGG16_bn(**kwargs):
    r"""VGG 16-layer model (configuration "D") with batch normalization
    `"Very Deep Convolutional Networks For Large-Scale Image Recognition" <https://arxiv.org/pdf/1409.1556.pdf>`_
    """
    return _vgg('vgg16_bn', 'D', True, **kwargs)


def VGG19(**kwargs):
    r"""VGG 19-layer model (configuration "E")
    `"Very Deep Convolutional Networks For Large-Scale Image Recognition" <https://arxiv.org/pdf/1409.1556.pdf>`_
    """
    return _vgg('vgg19', 'E', False, **kwargs)


def VGG19_bn(**kwargs):
    r"""VGG 19-layer model (configuration 'E') with batch normalization
    `"Very Deep Convolutional Networks For Large-Scale Image Recognition" <https://arxiv.org/pdf/1409.1556.pdf>`_
    """
    return _vgg('vgg19_bn', 'E', True, **kwargs)

3.模型测速代码：

import argparse
import os
import re
import subprocess
import time
import models.resnet as res_models
import models.vgg as vgg_models
import numpy as np
from jittor import nn
import jittor as jt
jt.flags.use_cuda = 1


def parse_args():
    parser = argparse.ArgumentParser()
    parser.add_argument('-a', '--arch', default='ResNet18', type=str)

    parser.add_argument('--batch-size', type=int, default=64)  # 128
    parser.add_argument('--learning-rate', type=float, default=0.025)
    parser.add_argument('--momentum', type=float, default=0.9)
    parser.add_argument('--weight-decay', type=float, default=1e-4)

    parser.add_argument('--dynamic', action='store_true')
    parser.add_argument('--single', action='store_false')
    parser.add_argument('--compiled', action='store_true')
    parser.add_argument('--step', type=int, default=100)

    return parser.parse_args()


def main():
    args = parse_args()
    worker(args)

def worker(args):
    if 'ResNet' in args.arch:
        model = res_models.__dict__[args.arch]()
    elif 'VGG' in args.arch:
        model = vgg_models.__dict__[args.arch]()
    else:
        raise NotImplementedError

    def train_func():
        model.train()

        logits = model(image)
        loss = nn.cross_entropy_loss(logits, label, ignore_index=255)
        optimizer.step(loss)

    def eval_func():
        model.eval()
        logits = model(image)


    image = np.random.random([args.batch_size, 3, 224, 224]).astype(np.float32)
    label = np.random.randint(1000, size=[args.batch_size , 1])
    image = jt.float32(image)
    label = jt.int64(label)
    optimizer = nn.SGD(model.parameters(), args.learning_rate, args.momentum, args.weight_decay)


    for i in range(10):
        train_func()

    train_start_time = time.time()
    for i in range(args.step):
        train_func()

    train_time = (time.time() - train_start_time) / args.step

    eval_start_time = time.time()
    for i in range(args.step):
        eval_func()

    eval_time = (time.time() - eval_start_time) / args.step
    stdout = subprocess.getoutput('nvidia-smi')
    mem = re.findall(r'\|  (.*?)MiB /', stdout)[0].strip()

    print('Jittor, Model:{0},Dy:{1},#GPU:{2},batch_size:{3},mem:{4}M, avg_train_time:{5:.3f}ms, avg_eval_time:{6:.3f}ms'. \
        format(args.arch, args.dynamic, 1, args.batch_size, mem, train_time * 1000, eval_time * 1000))


if __name__ == "__main__":
    main()

模型测速

• 测试设置：batch_size=64, GPU:1张1080ti, 测试迭代次数: 100;

  • 测试模型代码由jittor repo给出,
  • 训练和预测代码根据jittor exmaple写出。

• 测试结果

|Model | avg training time/iter | avg evaluate time/iter| |------- |--- | --- | ResNet-18 | 107.17ms | 1.26ms VGG-16 | 1.44ms |0.44ms

• 从测试结果看，提出以下问题： 1.ResNet和VGG都存在inference time非常短，其他深度学习框架约几十毫秒。请问是上述测试代码中，对于jittor的使用存在问题么？ 2.VGG-16的每个iter训练时间也非常短，极不正常。 期望得到作者的解答，非常感谢。

错误

新建了一个conda环境后使用pip安装，测试安装，出现以下错误：

暂时解决

如果使用系统自带的python和pip安装就没有问题，据悉是conda环境中修改了LD_LIBRARY_PATH导致无法找到glibcxx。 由于对自环境有刚需，所以请教解决方法，或加入conda install支持。

环境

system: openSUSE Tumbleweed (latest 5.16) compiler: gcc/g++11.2.1 cuda/nvcc: 11.6 with update 1 python: 3.8.12 conda 4.11.0/4.8.3

同时运行两个代码的时候第二个代码会报错：

RuntimeError: Wrong inputs arguments, Please refer to examples(help(jt.sync)).

Types of your inputs are: self = Var, args = (),

The function declarations are: void sync(bool device_sync = false)

Failed reason:[f 0909 17:29:40.576724 60 executor.cc:527] Execute fused operator(2324/4776) failed: [Op(0x5623cd93d830:0:0:1:i1:o1:s0,broadcast_to->0x5623cd93d8e0),Op(0x5623cf951be0:0: 0:1:i1:o1:s0,broadcast_to->0x5623cf951c90),Op(0x5623cf9534b0:0:0:1:i2:o1:s0,binary.multiply->0x5623cf953540),Op(0x5623cfb976e0:0:0:1:i1:o1:s0,reindex_reduce.add->0x5623cf9565c0),]

Reason: [f 0909 17:29:39.839614 60 helper_cuda.h:126] CUDA error at ~/.cache/jittor/default/g++/jit/cudnn_conv_backward_x_Tx:float32__Ty:float32__Tw:float32__XFORMAT:abcd__WF ORMAT:oihw__YFOR...hash:74f24b7a5fa4fe17_op.cc:167 code=4( CUDNN_STATUS_INTERNAL_ERROR ) cudnnFindConvolutionBackwardDataAlgorithmEx( handle_, cudnnFdesc, w->ptr(), cudnnOdesc, y->ptr(), cudnnConvDesc, cudnnIdesc, x->ptr(), num_algos, &perf_count, perf_results, ws, max_ws_size) [e 0909 17:29:40.891578 60 helper_cuda.h:115] Peek CUDA error at ~/anaconda3/envs/py3/lib/python3.8/site-packages/jittor/src/mem/allocator/cuda_dual_allocator.h:101 code=700 ( cudaErrorIllegalAddress ) _cudaLaunchHostFunc(0, &to_free_allocation, 0) terminate called after throwing an instance of 'std::runtime_error'
what(): [f 0909 17:29:42.170272 60 helper_cuda.h:126] CUDA error at ~/anaconda3/envs/py3/lib/python3.8/site-packages/jittor/extern/cuda/cudnn/src/cudnn_warper.cc:34 code= 4( CUDNN_STATUS_INTERNAL_ERROR ) cudnnDestroy(cudnn_handle)

请问下是什么问题？

按照标准流程手动安装，输出如下:

> python -m jittor.test.test_example
[i 1121 15:17:16.379761 12 __init__.py:257] Found g++(10.2.0) at g++
[i 1121 15:17:16.420423 12 compiler.py:839] Jittor(1.2.1.1) src: /usr/lib/python3.8/site-packages/jittor
[i 1121 15:17:16.420647 12 compiler.py:840] cache_path: /home/peter/.cache/jittor/default/g++
[i 1121 15:17:16.427812 12 compiler.py:791] Found /usr/local/cuda/bin/nvcc(11.1.105) at /opt/cuda/bin/nvcc
[i 1121 15:17:16.522639 12 __init__.py:249] Found gdb(10.1) at /usr/bin/gdb.
[i 1121 15:17:16.529103 12 __init__.py:249] Found addr2line(2.35.1) at /usr/bin/addr2line.
[i 1121 15:17:16.627024 12 compiler.py:889] pybind_include: -I/usr/include/python3.8 -I/usr/lib/python3.8/site-packages/pybind11/include
[i 1121 15:17:16.677954 12 compiler.py:891] extension_suffix: .cpython-38-x86_64-linux-gnu.so
g++: error: /usr/lib/python3.8/site-packages/jittor/src/utils/cache_compile.cc: Not a directory
g++: error: /usr/lib/python3.8/site-packages/jittor/src/utils/log.cc: Not a directory
g++: error: /usr/lib/python3.8/site-packages/jittor/src/utils/tracer.cc: Not a directory
g++: error: /usr/lib/python3.8/site-packages/jittor/src/utils/jit_utils.cc: Not a directory
Traceback (most recent call last):
  File "/usr/lib/python3.8/runpy.py", line 185, in _run_module_as_main
    mod_name, mod_spec, code = _get_module_details(mod_name, _Error)
  File "/usr/lib/python3.8/runpy.py", line 111, in _get_module_details
    __import__(pkg_name)
  File "/usr/lib/python3.8/site-packages/jittor/__init__.py", line 16, in <module>
    from . import compiler
  File "/usr/lib/python3.8/site-packages/jittor/compiler.py", line 901, in <module>
    check_cache_compile()
  File "/usr/lib/python3.8/site-packages/jittor/compiler.py", line 775, in check_cache_compile
    recompile = compile(cc_path, cc_flags+f" {opt_flags} ", files, 'jit_utils_core'+extension_suffix, True)
  File "/usr/lib/python3.8/site-packages/jittor/compiler.py", line 69, in compile
    return do_compile(cmd)
  File "/usr/lib/python3.8/site-packages/jittor/compiler.py", line 47, in do_compile
    run_cmd(cmd)
  File "/usr/lib/python3.8/site-packages/jittor_utils/__init__.py", line 136, in run_cmd
    raise Exception(err_msg)
Exception: Run cmd failed: g++ /usr/lib/python3.8/site-packages/jittor/src/utils/cache_compile.cc /usr/lib/python3.8/site-packages/jittor/src/utils/log.cc /usr/lib/python3.8/site-packages/jittor/src/utils/tracer.cc /usr/lib/python3.8/site-packages/jittor/src/utils/jit_utils.cc   -Wall -Werror -Wno-unknown-pragmas -std=c++14 -fPIC -march=native  -fdiagnostics-color=always -I/usr/include/python3.8 -I/usr/lib/python3.8/site-packages/pybind11/include -I/usr/lib/python3.8/site-packages/jittor/src   -O2    -lstdc++ -ldl -shared  -o /home/peter/.cache/jittor/default/g++/jit_utils_core.cpython-38-x86_64-linux-gnu.so

系统基本配置：

OS: Manjaro Linux x86_64 
Host: 80RQ Lenovo Rescuer-15ISK
Kernel: 5.9.8-2-MANJARO 

import os
import sys
import numpy as np
import jittor as jt
import jittor.models as jtmodels
import json
import time
jt.flags.use_cuda=1
a=np.random.rand(5, 3,512,512).astype('float32')
inputs=jt.array(a)
res=m(inputs)
t=time.time()
r=[]
for i in range(10):
    a=np.random.rand(5, 3,512,512).astype('float32')
    inputs=jt.array(a)
    res=m(inputs)
    r.append(res.max().numpy().tolist()[0])
print(time.time()-t)

在v100卡上对比当前版本的计图和pytorch1.4， 同样的程序计图平均在960ms，pytorch在950ms左右，感觉没有起到加速效果 同时计图占用了15516M显存，gc后会减少到6736M，显存占用较大， 请问针对这种推理方式有优化使用的方式么

I'm using module for package management on a cluster, the current environment is shown as following:

Currently Loaded Modulefiles:
  1) compiler/intel/2019.5.281    4) tools/mumps_seq/5.2.1        7) tools/matlab/R2017a
  2) mpi/intel/2019.5.281         5) tools/superlu/5.2.1          8) compiler/cuda/10.1
  3) tools/tmux/1.8               6) tools/suitesparse/5.6.0      9) interpreter/python/3.7-gpu

Note that python 3.7 and nvcc 10.1 are both loaded, and currently c/c++ compiler is icc, so let's start jittor program

[i 0518 20:13:45.097043 52 compiler.py:851] Jittor(1.2.3.8) src: /home/users/$MYNAME$/.local/lib/python3.7/site-packages/jittor
[i 0518 20:13:45.097205 52 compiler.py:852] g++ at /software/local/bin/g++
[i 0518 20:13:45.097302 52 compiler.py:853] cache_path: /home/users/$MYNAME$/.cache/jittor/default/g++
which: no nvcc in (/usr/local/cuda/bin)
[i 0518 20:13:45.195274 52 __init__.py:255] Found gdb(7.6.1) at /bin/gdb.
[i 0518 20:13:45.212988 52 __init__.py:255] Found addr2line(7.3) at /bin/addr2line.
[i 0518 20:13:45.338448 52 compiler.py:893] py_include: -I/software/python/virtualenv/py3.7-gpu/include/python3.7m -I/software/python/virtualenv/py3.7-gpu/include/python3.7m
[i 0518 20:13:45.437864 52 compiler.py:895] extension_suffix: .cpython-37m-x86_64-linux-gnu.so

I don't know why jittor is trying to obtain g++, which i haven't loaded (so the environment is not correct for g++) and why it hasn't found nvcc, maybe compilers should be found by using current environment path

最近使用jittor读取torch嵌套pth文件发现这样一个问题：一旦torch的sequential或者嵌套module带有名字，jittor就无法读取（和jittor sequential数字索引有关。）因此用户必须按照数字顺序命名。我自己尝试使用回溯来是的用户可以任意命名。 另外还有一个小bug：打印module时，其中的LeakyReLU层会显示MakeModule，查阅文档似乎ReLU系层都存在这个问题。

Minimal Reproduce

import jittor
a = jittor.ones((2,2))
b = jittor.ones((2,2))
jt.linalg.einsum('ac,     ac -> a', a, b) # <- one space is enough to cause the error
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/zpool-portable/shared/diffusion/diffusion-pytorch/lib/python3.10/site-packages/jittor/linalg.py", line 497, in einsum
    output_shape = [int(x) for x in einsum_outshape(string, args)]
  File "/zpool-portable/shared/diffusion/diffusion-pytorch/lib/python3.10/site-packages/jittor/linalg.py", line 495, in einsum_outshape
    return tuple(shps[(np_cpu.concatenate(inop[:-1])[:,None]==inop[-1]).argmax(0)].astype(np_cpu.int64))
IndexError: index 4 is out of bounds for axis 0 with size 4

Expected behavior

The corresponding numpy einsum func has no such issue,

c = numpy.ones((2,2))
d = numpy.ones((2,2))
nump.einsum('a    c   , a c ->  a ', c , d) # <-whatever space has no effect

perhaps, need to strip any space in einsum_outshape or split by space?

Traceback (most recent call last): File "main.py", line 6, in jt.flags.use_cuda = 1 RuntimeError: Wrong inputs arguments, Please refer to examples(help(jt.Flags.use_cuda)).

Types of your inputs are: self = Flags, arg = int,

The function declarations are: void _set_use_cuda(int v) void _set_use_cuda(bool v)

Failed reason:[f 1219 15:54:51.134013 56 cuda_flags.cc:44] Check failed: value==0 No CUDA found.

Patching CVE-2007-4559

Hi, we are security researchers from the Advanced Research Center at Trellix. We have began a campaign to patch a widespread bug named CVE-2007-4559. CVE-2007-4559 is a 15 year old bug in the Python tarfile package. By using extract() or extractall() on a tarfile object without sanitizing input, a maliciously crafted .tar file could perform a directory path traversal attack. We found at least one unsantized extractall() in your codebase and are providing a patch for you via pull request. The patch essentially checks to see if all tarfile members will be extracted safely and throws an exception otherwise. We encourage you to use this patch or your own solution to secure against CVE-2007-4559. Further technical information about the vulnerability can be found in this blog.

If you have further questions you may contact us through this projects lead researcher Kasimir Schulz.

Describe the bug

RuntimeError occurs when calling jt.matmul in WSL on an NVIDIA GeForce MX450 laptop.

Full Log

[i 1031 23:09:50.689249 80 cuda_flags.cc:32] CUDA enabled.
---------------------------------------------------------------------------
RuntimeError                              Traceback (most recent call last)
<ipython-input-3-b8f9d3eceb14> in <module>
----> 1 jt.flags.use_cuda = 1; jt.matmul(jt.rand((1, 1, 100, 100)), jt.rand(((1, 16, 100, 1)))).shape

~/anaconda3/envs/gm-jittor/lib/python3.7/site-packages/jittor/nn.py in matmul(a, b)
    122             # a: [..., n, m], b: [..., m, k], c:[..., n, k]
    123             if jt.flags.use_cuda and jt.compile_extern.cublas_ops:
--> 124                 return jt.compile_extern.cublas_ops.cublas_batched_matmul(a, b, 0, 0)
    125         shape = []
    126         len_c = max(len_a, len_b)

RuntimeError: Wrong inputs arguments, Please refer to examples(help(jt.cublas_batched_matmul)).

Types of your inputs are:
 self   = module,
 args   = (Var, Var, int, int, ),

The function declarations are:
 VarHolder* cublas_batched_matmul(VarHolder* a, VarHolder* b,  bool trans_a,  bool trans_b)

Failed reason:[f 1031 23:09:50.690284 80 cublas_batched_matmul_op.cc:75] Check failed a->shape[i](1) == b->shape[i](16) Something wrong ... Could you please report this issue?

Minimal Reproduce

# On CPU, jt.matmul() works
jt.flags.use_cuda = 0; jt.matmul(jt.rand((1, 1, 100, 100)), jt.rand(((1, 16, 100, 1)))).shape
# On GPU, it doesn`t work
jt.flags.use_cuda = 1; jt.matmul(jt.rand((1, 1, 100, 100)), jt.rand(((1, 16, 100, 1)))).shape

Possible Solution

# Add repeat(1, 16, 1, 1) to explicitly specify the shape to broadcast
jt.flags.use_cuda = 1; jt.matmul(jt.rand((1, 1, 100, 100)).repeat(1, 16, 1, 1), jt.rand(((1, 16, 100, 1)))).shape

However, this solution results in large computational error.

a = torch.randn(1,1,100,100).numpy()
b = torch.randn(1,16,100,1).numpy()
c_torch = torch.matmul(torch.tensor(a), torch.tensor(b)).numpy()
c_jt = jt.matmul(jt.Var(a).repeat(1, 16, 1, 1), jt.Var(b)).numpy()
np.testing.assert_almost_equal(c_torch, c_jt) # Passed

jt.flags.use_cuda = 1
c_jt = jt.matmul(jt.Var(a).repeat(1, 16, 1, 1), jt.Var(b)).numpy()
np.testing.assert_almost_equal(c_torch, c_jt) # AssertionError

AssertionError:
Arrays are not almost equal to 7 decimals

Mismatched elements: 1350 / 1600 (84.4%)
Max absolute difference: 1.1444092e-05
Max relative difference: 3.5474255e-05
 x: array([[[[  4.539161 ],
         [  5.9756746],
         [ 13.752586 ],...
 y: array([[[[  4.53916  ],
         [  5.9756722],
         [ 13.752583 ],...

Expected behavior

jt.matmul should work correctly on CUDA.

Describe the bug

Mac import nerf 或运行python tools/run_net.py --config-file ./projects/ngp/configs/ngp_base.py --task test 的时候会崩溃

Full Log

[i 1022 00:01:50.683523 00 compiler.py:955] Jittor(1.3.5.21) src: /Users/mei/opt/anaconda3/lib/python3.7/site-packages/jittor [i 1022 00:01:50.734341 00 compiler.py:956] g++ at /usr/bin/g++(12.0.0) [i 1022 00:01:50.734457 00 compiler.py:957] cache_path: /Users/mei/.cache/jittor/jt1.3.5/g++12.0.0/py3.7.6/Darwin-20.6.0-x92/IntelRCoreTMi5x84/default [i 1022 00:01:51.089361 00 init.py:232] using 4 procs for compiling. [i 1022 00:01:51.343270 00 jit_compiler.cc:28] Load cc_path: /usr/bin/g++ clang: error: unsupported option '--showme:version' [i 1022 00:01:52.669768 00 compile_extern.py:517] mpicc not found, distribution disabled. Caught segfault at address 0x0, thread_name: '', flush log... [bt] Execution path: [bt] #1 1 jit_utils_core.cpython-37m-darwin.s 0x00000001058371b8 _ZN6jittor18segfault_sigactionEiP9__siginfoPv + 2216 [bt] #2 2 libsystem_platform.dylib 0x00007fff20821d7d _sigtramp + 29 [bt] #3 3 python 0x0000000104ea0cd1 _PyObject_Malloc + 17 [bt] #4 4 libc++abi.1.0.dylib 0x00000001059eebd8 __cxa_throw + 104 [bt] #5 5 jittor_core.cpython-37m-darwin.so 0x0000000105a0a671 _ZN6jittor15LogFatalVoidifyaaERNS_3LogE + 129 [bt] #6 6 jittor_core.cpython-37m-darwin.so 0x0000000105c1598d _ZN6jittor15setter_use_cudaEi + 253 [bt] #7 7 jittor_core.cpython-37m-darwin.so 0x0000000105c15880 _ZN6jittor12set_use_cudaERKi + 16 [bt] #8 8 jittor_core.cpython-37m-darwin.so 0x0000000105b9eb06 _ZZN6jittor18pyjt_def_jit_flagsEP7_objectEN5$_1158__invokeES1_S1_Pv + 374 [bt] #9 9 python 0x0000000104e9ef4b _PyObject_GenericSetAttrWithDict + 219 [bt] #10 10 python 0x0000000104e9dd40 PyObject_SetAttr + 96 [bt] #11 11 python 0x0000000104f76dfb _PyEval_EvalFrameDefault + 30555 [bt] #12 12 python 0x0000000104f6e46e _PyEval_EvalCodeWithName + 414 [bt] #13 13 python 0x0000000104f68e9b builtin_exec + 347 [bt] #14 14 python 0x0000000104e3ec0f _PyMethodDef_RawFastCallDict + 255 [bt] #15 15 python 0x0000000104e4031d PyCFunction_Call + 61 Segfault, exit

Minimal Reproduce

Reproduce this error with a file or several lines of code. If it is not possible, leave it blank.

Expected behavior

A clear and concise description of what you expected to happen.

If you are submitting an issue for the first time, please refer to our guideline

Describe the bug

在使用nn.CrossEntropyLoss()时，指定weight参数为jitter.ones(num_classes, dtype='float32')和置为None的运行结果不同。且指定包括全一在内的值后会带来梯度爆炸。

Full Log

这是指定了weight具体值的训练曲线。

这是不指定weight值的训练曲线。

Minimal Reproduce

数据集是CIFAR-10，简单的Conv+Linear，类似于Jittor教程的MNIST分类网络。

Expected behavior

希望两者训练行为至少一致，最好是都不会发生梯度爆炸。