rynmurdock's picture
init
c5ca37a
description: Train VQ on Bird Dataset
auth:
# which virtual cluster you belong to (msrlabs, etc.). Everyone has access to "pnrsy".
vc: msrlabspvc12 # msrlabs
# physical cluster to use (cam, gcr, rr1) or Azure clusters (eu1, eu2, etc.)
# cluster: rr2, eu2, eu1 et1
cluster: eu2
# docker environment (vm) in which your job will run. we provide "generic" dockers
# with the main deep learning toolkit installed (PyTorch, TF, Chainer, etc.)
docker:
# image: philly/jobs/custom/generic-docker:py27
# registry: phillyregistry.azurecr.io
image: vlnres/vqvae:v1 # chunyl/vqvae:v2
registry: index.docker.io
storage:
_default:
#use_phillyfs: True
storage_account_name: sslm
container_name: vqvae
mount_path: /mnt/_default
code:
# local directory of the code. this will be uploaded to the server.
# $CONFIG_DIR is expanded to the directory of this config file
code_upload: False
remote_dir: vq-vae-2-pytorch/
local_dir: $CONFIG_DIR/src
#data:
# data upload is not required for this example
#data_upload: False
search:
job_template:
name: vq_{experiment_name:s}_{image_size_option:.1f}
sku: G4 # G4 # G1
command:
- python train_vqvae.py --philly --dataset_name bird --size {image_size_option} --batch 512
max_trials: 20
type: grid
params:
- name: image_size_option
spec: discrete
values: [64,128] # [top,bottom]