Tensorflow benchmark#8522
Conversation
… tensorflowbenchmark
… tensorflowbenchmark
benchmark/cluster/vgg16/vgg16_tf.py
Outdated
| print("Pass = %d, Train speed = %f imgs/s, Test accuracy = %f\n" % | ||
| (pass_id, num_samples / train_elapsed, pass_test_acc)) | ||
|
|
||
| sv.stop() |
There was a problem hiding this comment.
There was a problem hiding this comment.
It comes from other examples and should be cleaned up.
Done.
benchmark/cluster/vgg16/Dockerfile
Outdated
| #FROM python:2.7.14 | ||
| FROM nvidia/cuda:8.0-cudnn5-runtime-ubuntu16.04 | ||
| RUN apt-get update && apt-get install -y python | ||
| FROM paddlepaddle/paddle:latest-dev |
There was a problem hiding this comment.
Do not use this image as base image, CI built image does not enable WITH_DISTRIBUTED yet.
There was a problem hiding this comment.
This not product image, it's the develop image.
There was a problem hiding this comment.
I see, but no need, paddlepaddle/paddle:latest-dev is rather large.
|
|
||
| # NOTE: By default CI built wheel packages turn WITH_DISTRIBUTE=OFF, | ||
| # so we must build one with distribute support to install in this image. | ||
| RUN pip install paddlepaddle |
There was a problem hiding this comment.
The whl package may need to change a lot for test, but dataset download is slow and no need to change. Use RUN pip install paddlepaddle to download dataset first.
There was a problem hiding this comment.
8 # if download slowly, you can add proxy here.
9 # ENV https_proxy=
This can be confusing
The whole package may need to change a lot for test, but dataset download is slow and no need to change. Use RUN pip install paddlepaddle to download dataset first.
Do we need to add dataset from local?
There was a problem hiding this comment.
I also added downloading "flowers" dataset, is so large that we only want to download it once.
| value: "/workspace" | ||
| - name: PADDLE_INIT_PORT | ||
| value: "30236" | ||
| - name: PORT |
There was a problem hiding this comment.
Should change vgg16_fluid.py to use PADDLE_INIT_PORT rather than add a new env.
2 participants
No description provided.