mlcommons / training

Reference implementations of MLPerf™ training benchmarks
https://mlcommons.org/en/groups/training
Apache License 2.0
1.57k stars 548 forks source link

Correct version of Tensorflow for minigo #535

Closed mahmoodn closed 1 year ago

mahmoodn commented 2 years ago

Hi Installing and running workloads take more time than my expectation. It seems that minigo has been written with Tensorflow 1.15. Is that right? On a container with Ubuntu 18.04 and CUDA 11.2, I get this error

Copying /research/results/minigo/final/estimator_working_dir/model.ckpt-1.data-00000-of-00001 to /research/results/minigo/final/models/000000-bootstrap.data-00000-of-00001
Copying /research/results/minigo/final/estimator_working_dir/model.ckpt-1.meta to /research/results/minigo/final/models/000000-bootstrap.meta
Transform /research/results/minigo/final/models/000000-bootstrap.index
python3: can't open file '/usr/local/lib/python3.6/dist-packages/tensorflow_core/python/keras/api/_v1/python/tools/freeze_graph.py': [Errno 2] No such file or directory

More information about packages:

# pip3 list | grep -i keras
Keras-Applications             1.0.8
Keras-Preprocessing            1.1.2
# pip3 list | grep tensor
tensorboard                    1.15.0
tensorflow-estimator           1.15.1
tensorflow-gpu                 1.15.0

How can I fix the problem?

johntran-nv commented 1 year ago

Closing since minigo is retired.