InteractiveErrors/gpt2_rl_hyperparameter_search.yaml at master · QMrpy/InteractiveErrors · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
description: Hyperparameter Search for RL on Graph (H2)

target:
  service: amlk8s
  name: itpeastusv100cl2
  vc: resrchvc

environment:
  image: pytorch/pytorch:1.6.0-cuda10.1-cudnn7-devel
  setup:
    - pip install --user numpy
    - pip install --user networkx
    - pip install --user numba
    - pip install --user stable-baselines3==1.0
    - pip install --user transformers
    - pip install --user gym
    - pip install --user sklearn
    - pip install --user pandas
    - pip install --user sentencepiece
    - pip install --user spacy==3.0.3
    - pip install --user https://github.com/explosion/spacy-models/releases/download/en_core_web_lg-3.0.0/en_core_web_lg-3.0.0-py3-none-any.whl

code:
  local_dir: $CONFIG_DIR/h2_code

data:
  local_dir: $CONFIG_DIR/h2_data_models/
  remote_dir: h2_data_models

search:
    job_template:
        name: search_gpt2_turing_{auto:s}
        sku: G1
        command:
        - python graph_rl.py --input_file $$AMLT_DATA_DIR/experiments_data/turing_leakages
          --output_file $$AMLT_OUTPUT_DIR/h2_turing_gpt2 --cls_model $$AMLT_DATA_DIR/roberta
          --gen_model $$AMLT_DATA_DIR/gpt2_medium --batch_size 64 --top_k {top_k} --steps 5000
          --horizon {horizon} --sim_threshold {sim_threshold}
    type: grid
    max_trials: 16
    params:
        - name: sim_threshold
          spec: discrete
          values: [0.5, 0.6, 0.7, 0.8]
        - name: top_k
          spec: discrete
          values: [200, 300]
        - name: horizon
          spec: discrete
          values: [200, 300]