protnote/hyperdrive_seed_replicates.yml at main · microsoft/protnote · 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
53
54
55
56
57
description: Protein Functions replicate experiment with 5 different seeds and 3 different weight sampling schemes.
  Up to 46 epochs.

target:
  service: sing
  name: barlow01
  workspace_name: bio0-ext

environment:
  image: amlt-sing/acpt-2.0.1-py3.10-cuda11.8 #azureml/openmpi4.1.0-cuda11.8-cudnn8-ubuntu22.04:latest
  conda_yaml_file: $CONFIG_DIR/environment.yml
  skip_conda_packages_on_sing:
  - tensorflow
  - cudatoolkit
  - deepspeed
  - pip
  - python\b

code:
  local_dir: $CONFIG_DIR/

data:
  local_dir: $CONFIG_DIR/data
  remote_dir: data

# SKU usage: G1 (single GPU), G4 (quad GPU), G4-V100 (1 machine, 4 V100 gpus), etc...
search:
  job_template:
    name: "{experiment_name}_SEED_{SEED}_SEQUENCE_WEIGHT_AGG_{SEQUENCE_WEIGHT_AGG}"
    sku: G8-V100
    priority: high
    preemptible: False
    command:
    - python main.py
      --nodes 1
      --gpus 8
      --train-path-name TRAIN_DATA_PATH
      --validation-path-name VAL_DATA_PATH
      --name "{experiment_name}_{SEED}_{SEQUENCE_WEIGHT_AGG}"
      --override ESTIMATE_MAP True
      --amlt
      --use-wandb
    submit_args:
      env:
        WANDB_BASE_URL: "https://microsoft-research.wandb.io"
        WANDB_API_KEY: "$WANDB_API_KEY"
  type: hyperdrive
  sampling: grid
  max_trials: 10
  parallel_trials: 10
  params:
    - name: SEED
      values: [12,22,32,42,52]
    - name: SEQUENCE_WEIGHT_AGG
      values: ['sum','mean']