Skip to first unread message
Guy N
Feb 23, 2017, 3:30:33 AM2/23/17
to Caffe Users
I saw that recently a new commit was inserted to Caffe, that should enable multi-GPU training with Python interface/Python custom layers.
I can't seem to train a network using multiple-GPU from my Python code.
I cloned the updated repository and built Caffe with USE_NCCL on.
The commands I tried to use in my Python script:
caffe.set_mode_gpu()
caffe.set_solver_count(2)
The error I get:
> F0223 10:27:41.249380 23468 python_layer.hpp:25] PythonLayer does not
> support CLI Multi-GPU, use train.py
When running the script from the command line, it simply uses 1 GPU only.
I can't seem to train a network using multiple-GPU from my Python code.
I cloned the updated repository and built Caffe with USE_NCCL on.
The commands I tried to use in my Python script:
caffe.set_mode_gpu()
caffe.set_solver_count(2)
The error I get:
> F0223 10:27:41.249380 23468 python_layer.hpp:25] PythonLayer does not
> support CLI Multi-GPU, use train.py
When running the script from the command line, it simply uses 1 GPU only.
Reply all
Reply to author
Forward
0 new messages