File size: 1,585 Bytes
3a8e4d3 | 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 | # @package _global_
# All parameters below will be merged with parameters from default configurations set above
# this allows you to overwrite only specified parameters
# Options for training:
# - /data: [pdb, pdb_sp, pdb_sp_afesm, pdb_sp_afesme]
# - /model/architecture:
# [foldingdit_100M, foldingdit_360M, foldingdit_700M,
# foldingdit_1.1B, foldingdit_1.6B, foldingdit_3B]
# - /trainer: [default, fsdp]
defaults:
- override /data: pdb
- override /model: simplefold
- override /model/architecture: foldingdit_100M
- override /model/sampler: euler_maruyama
- override /model/processor: protein_processor
- override /callbacks: default
- override /trainer: default
- override /logger: tensorboard
# load_ckpt_path: [PATH_TO_YOUR_CKPT] # uncomment to load a checkpoint
seed: 12345
trainer:
max_steps: 300000
val_check_interval: 10000 # this controls BOTH checkpoint steps and validation
accumulate_grad_batches: 1 # gradient accumulation
data:
batch_size: 1 # should be the same as the number of GPUs
num_workers: 16
model:
_target_: models.simplefold.simplefold.SimpleFold
ema_decay: 0.999
clip_grad_norm_val: 2.0
esm_model: "esm2_3B"
lddt_weight_schedule: False # set to True in finetuning phase
scheduler:
_target_: onescience.utils.simplefold.lr_scheduler.LinearWarmup
_partial_: true
min_lr: 1e-6
max_lr: ${model.optimizer.lr}
warmup_steps: 5000
architecture:
esm_model: ${model.esm_model}
processor:
scale: 16.0
ref_scale: 5.0
multiplicity: 16 # number of copies per GPU
|