Skip to content

memory leak from boosters with CUDA #207

@Moelf

Description

@Moelf
[ Info: Training machine(XGBoostClassifier(test = 1, …), …).
[ Info: XGBoost: starting training.
┌ Warning: [04:42:46] WARNING: [/workspace/srcdir/xgboost/src/common/error_msg.cc:27](https://jiling-notebook-1.notebook.af.uchicago.edu/workspace/srcdir/xgboost/src/common/error_msg.cc#line=26): The tree method `gpu_hist` is deprecated since 2.0.0. To use GPU training, set the `device` parameter to CUDA instead.
│ 
│     E.g. tree_method = "hist", device = "cuda"
└ @ XGBoost ~/.julia/packages/XGBoost/nqMqQ/src/XGBoost.jl:34
┌ Error: Problem fitting the machine machine(XGBoostClassifier(test = 1, …), …). 
└ @ MLJBase ~/.julia/packages/MLJBase/7nGJF/src/machines.jl:694
[ Info: Running type checks... 
[ Info: Type checks okay. 

XGBoostError: (caller: XGBoosterUpdateOneIter)
[04:42:46] [/workspace/srcdir/xgboost/src/tree/updater_gpu_hist.cu:781](https://jiling-notebook-1.notebook.af.uchicago.edu/workspace/srcdir/xgboost/src/tree/updater_gpu_hist.cu#line=780): Exception in gpu_hist: [04:42:46] [/workspace/srcdir/xgboost/src/c_api/../data/../common/device_helpers.cuh:431](https://jiling-notebook-1.notebook.af.uchicago.edu/workspace/srcdir/xgboost/src/common/device_helpers.cuh#line=430): Memory allocation error on worker 0: Caching allocator
- Free memory: 12582912
- Requested memory: 9232383

this happens with less and less Free memory: so something is leaking memory... I'm using XGBoost via MLJ, any clue?

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions