Where communities thrive


  • Join over 1.5M+ people
  • Join over 100K+ communities
  • Free without limits
  • Create your own community
People
Repo info
Activity
    1575018690
    @1575018690
    After installing nni, run the mnist sample program, open the web ui, there is an error. How to solve the error in the picture?
    the error INFO (nni.runtime.msg_dispatcher_base/MainThread) Dispatcher started. After installing nni, run the mnist sample program, open the web ui, there is an error. How to solve the error?
    1575018690
    @1575018690
    image.png
    1575018690
    @1575018690
    image.png
    How to solve this problem?
    kvartet
    @kvartet
    @1575018690 From your error message, this is because you do not have permission to this directory. You can configure logDir to a folder with permission, or start your experiment in administrator mode. If it doesn’t work, please open an issue with your nnimanager.log in NNI Github community.
    1575018690
    @1575018690
    @kvartet thanks
    If you use nni to adjust the hyper-parameters, what server is better, I have no experience in using the server to calculate the parameters
    oshritshtossel
    @oshritshtossel
    Hi,
    I'm trying to use Retiarii Experiment with pytorch lightning as shown here: https://nni.readthedocs.io/en/latest/NAS/retiarii/WriteTrainer.html
    But I crash in a json dump process with this error: TypeError: Object of type <class 'torch.dtype'> could not be encoded by TricksEncoder using encoders [...]. You can add an encoders for this type using extra_obj_encoders. If you want to 'skip' this object, consider using fallback_encoders like str or lambda o: None.
    kvartet
    @kvartet

    But I crash in a json dump process with this error: TypeError: Object of type <class 'torch.dtype'> could not be encoded by TricksEncoder using encoders [...]. You can add an encoders for this type using extra_obj_encoders. If you want to 'skip' this object, consider using fallback_encoders like str or lambda o: None.

    In order to track your questions and help others who have the same questions as you, can you open an issue(https://github.com/microsoft/nni/issues/new/choose) in our Github repo, thanks~

    kvartet
    @kvartet

    If you use nni to adjust the hyper-parameters, what server is better, I have no experience in using the server to calculate the parameters

    Any server is fine.

    1575018690
    @1575018690
    image.png
    When I set gpunum to 1, trial operation reports an error
    image.png
    May I ask what caused this problem?
    kvartet
    @kvartet
    @1575018690 please set useActiveGpu=True(https://nni.readthedocs.io/en/latest/Tutorial/ExperimentConfig.html#useactivegpu) and try again. If it doesn't work, please open an issue in NNI Github repo(https://github.com/microsoft/nni/issues/new/choose).
    1575018690
    @1575018690
    nnierror1.png
    nnierror2.png
    system:window10
    pytorch
    Why is the default metric not displayed? However, the value of metric exists
    1575018690
    @1575018690
    nnierror3.png
    nnierror4.png
    When I change gpuNum=1 to 0, the default matric can be displayed normally. What is the reason?
    kvartet
    @kvartet
    just to record, above issues have been tracked in NNI Github repo: microsoft/nni#3610 & microsoft/nni#3609
    Dannypung
    @Dannypung
    parser.add_argument('--tolerance',type=int,default=100,help='tolerance for earlystopping')
    If my parameters are set this way, can I search for parameters directly as shown in the example
    Will Sackfield
    @8W9aG
    does NNI have a solution for DL that allows me to feed in the inputs, outputs and evaluate the model while leaving me out of any decisions regarding the neural architecture of the model? Most of the examples I have seen force me to make some layer choices
    Shadow Walker
    @shadowwalker2718
    I am trying to run NNI in remote servers but got this error:
    Error: Cannot find module '/Users/pmg/git.repo/nni/ts/nni_manager/dist/package.json'
        at Function.Module._resolveFilename (internal/modules/cjs/loader.js:636:15)
        at Function.Module._load (internal/modules/cjs/loader.js:562:25)
        at Module.require (internal/modules/cjs/loader.js:692:17)
        at require (internal/modules/cjs/helpers.js:25:18)
        at Promise.resolve.then (/Users/pmg/git.repo/nni/ts/nni_manager/dist/common/utils.js:276:34)
        at process._tickCallback (internal/process/next_tick.js:68:7)
    Anybody knows what I missed?
    kvartet
    @kvartet

    parser.add_argument('--tolerance',type=int,default=100,help='tolerance for earlystopping')
    If my parameters are set this way, can I search for parameters directly as shown in the example

    You can overwrite the args with the parameters got from nni.get_next_parameter(). If you still have questions, we can discuss in the https://github.com/microsoft/nni/discussions/3747

    kvartet
    @kvartet

    does NNI have a solution for DL that allows me to feed in the inputs, outputs and evaluate the model while leaving me out of any decisions regarding the neural architecture of the model? Most of the examples I have seen force me to make some layer choices

    hello @8W9aG, I open an issue microsoft/nni#3830 in the Github repo, let's discuss it there.

    I am trying to run NNI in remote servers but got this error:

    Error: Cannot find module '/Users/pmg/git.repo/nni/ts/nni_manager/dist/package.json'
        at Function.Module._resolveFilename (internal/modules/cjs/loader.js:636:15)
        at Function.Module._load (internal/modules/cjs/loader.js:562:25)
        at Module.require (internal/modules/cjs/loader.js:692:17)
        at require (internal/modules/cjs/helpers.js:25:18)
        at Promise.resolve.then (/Users/pmg/git.repo/nni/ts/nni_manager/dist/common/utils.js:276:34)
        at process._tickCallback (internal/process/next_tick.js:68:7)

    copy ./nni/ts/nni_manager/package.json to ./nni/ts/nni_manager/dist/package.json may solve your issue.
    And recommend upgrade nni repo to the latest version, this is a bug in the old version.

    kalavinka
    @kalavinka
    Hello,
    How can I force nni to start from a given point (set of parameters) in the search space?
    kvartet
    @kvartet

    Hello,
    How can I force nni to start from a given point (set of parameters) in the search space?

    hello @kalavinka, I open an issue microsoft/nni#3834 for you, let's discuss it there!

    ztzthu
    @ztzthu
    image.png
    Hello, I was trying to run nnictl experiment, but the webUI gave the following feedback: "Dispatcher stream error, tuner may have crashed." and failed immediately.
    I tried to look at the dispatcher log, but it always shows loading. what could be the problem ?
    kvartet
    @kvartet
    hello @ztzthu, you can open your nni-experiments folder in /home/userName/nni-experiments/experimentId to see the nnimanger.log and dispatcher.log. If this error message could not help you find the failed reason, you can open an issue in NNI Github Community with your log and config, etc.
    yuezhuang1387
    @yuezhuang1387
    image.png
    When I use NNI GUI, this problem appears. How can I solve it? I am very confused
    Patrick Lenihan
    @JorjiCostava_gitlab
    ERROR: Config V2 validation failed: ValueError('_AlgorithmConfig: Unrecognized fields builtinassessorname')
    ERROR: 'NoneType' object has no attribute 'get'
    Why do I get this error?
    Screenshot 2021-08-10 at 20.23.12.png
    Joshua Steier
    @Rock910

    TracerWarning: Converting a tensor to a Python boolean might cause the trace to be incorrect. We can't record the data flow of Python values, so this value will be treated as a constant in the future. This means that the trace might not generalize to other inputs!
    if x.shape[2] % self.stride == 0:
    /utils.py:41: TracerWarning: Converting a tensor to a Python boolean might cause the trace to be incorrect. We can't record the data flow of Python values, so this value will be treated as a constant in the future. This means that the trace might not generalize to other inputs!
    y.shape[0] <= x.shape[0] and y.shape[1] <= x.shape[1]
    utils.py:25: TracerWarning: Converting a tensor to a Python index might cause the trace to be incorrect. We can't record the data flow of Python values, so this value will be treated as a constant in the future. This means that the trace might not generalize to other inputs!
    x = x[:, :, crop_t:-crop_b, crop_l:-crop_r]

    WARNING (nni.compression.pytorch.utils.mask_conflict/MainThread) no multi-dimension masks found.
    [2021-08-15 06:34:55] INFO (nni.compression.pytorch.utils.mask_conflict/MainThread) detected conv prune dim: 0

    I get the above error when I use model speedup on my model, it hangs here and crashes.
    Any ideas?
    Thanks

    AnkitRai-22
    @AnkitRai-22
    Hello everyone!! I am willing to contribute to nni by working on "Migrating FBNet from NAS to Retiarii". It would be great if somebody could help me understand the current transfer process.
    wubo2180
    @wubo2180
    windows
    Junshan
    @EroData
    I met this problem, dose anyone know the solution?
    [2021-10-26 12:12:29] WARNING (GPUScheduler) gpu_metrics file does not exist!
    Junshan
    @EroData
    The wechat code is expired. Can anyone share the new?
    sdustwyx
    @sdustwyx
    Does NNI support multiple sets of super parameters to be input together