Spaces:
Runtime error
Runtime error
| description: Train VAE on Yelp Dataset | |
| auth: | |
| # which virtual cluster you belong to (msrlabs, resrchprojvc6, etc.). Everyone has access to "pnrsy". | |
| vc: resrchprojvc7 | |
| # physical cluster to use (cam, gcr, rr1) or Azure clusters (eu1, eu2, etc.) | |
| # cluster: rr2, eu2, eu1 et1 | |
| cluster: rr1 | |
| # 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: chunyl/pytorch-transformers:v0 | |
| registry: index.docker.io | |
| storage: | |
| _default: | |
| #use_phillyfs: True | |
| storage_account_name: textae | |
| container_name: bigtextae | |
| 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: code/ | |
| local_dir: $CONFIG_DIR/code | |
| #data: | |
| # data upload is not required for this example | |
| #data_upload: False | |
| search: | |
| job_template: | |
| name: exp_{experiment_name:s}_b{bs_option:.0f}_beta_{beta_option:.2f}_d_{dim_target_kl_option:.2f}_r0_{ratio_zero_option:.2f}_ra_{ratio_increase_option:.2f} | |
| sku: G4 # G4 # G1 | |
| command: | |
| - pip install --user --editable . | |
| - pip install --user azure | |
| - pip install --user tqdm | |
| - python examples/big_ae/run_lm_vae_training.py --use_philly --num_train_epochs 1.0 --beta {beta_option} --dim_target_kl {dim_target_kl_option} --ratio_zero {ratio_zero_option} --ratio_increase {ratio_increase_option} --dataset Yelp --per_gpu_train_batch_size {bs_option} --per_gpu_eval_batch_size 1 --block_size 300 --output_dir ../output/philly_vae_yelp_b{beta_option}_d{dim_target_kl_option}_r0{ratio_zero_option}_ra{ratio_increase_option} --encoder_model_type bert --encoder_model_name_or_path bert-base-cased --decoder_model_type gpt2 --decoder_model_name_or_path gpt2 --do_train --train_data_file ../data/datasets/yelp_data/train.txt --do_eval --eval_data_file ../data/datasets/yelp_data/test.txt --overwrite_output_dir --save_steps 2000 --logging_steps 100 --use_pretrained_model --use_pretrained_vae --checkpoint_dir ../output/philly_rr3scl_g8_vae_wikipedia_pretraining_beta_schedule_beta1.0_d1.0_ro0.5_ra0.25 --gloabl_step_eval 760000 | |
| max_trials: 50 | |
| type: grid | |
| params: | |
| - name: bs_option | |
| spec: discrete | |
| values: [4] # | |
| - name: beta_option | |
| spec: discrete | |
| values: [1.0] # | |
| - name: dim_target_kl_option | |
| spec: discrete | |
| values: [0.01,0.05,0.25,0.5,1.0] # | |
| - name: ratio_zero_option | |
| spec: discrete | |
| values: [0.5] # | |
| - name: ratio_increase_option | |
| spec: discrete | |
| values: [0.25] # | |
| # - name: bs_option | |
| # spec: discrete | |
| # values: [4] # | |
| # - name: beta_option | |
| # spec: discrete | |
| # values: [0.25,1.0] # | |
| # - name: dim_target_kl_option | |
| # spec: discrete | |
| # values: [0.01,0.05,0.25,0.5,1] # | |
| # - name: ratio_zero_option | |
| # spec: discrete | |
| # values: [0.5] # | |
| # - name: ratio_increase_option | |
| # spec: discrete | |
| # values: [0.25] # | |