Number of Optuna trials: 25
Best trial objective: 0.5586 (best trial number: 10)
Best hyperparameters:
- learning_rate: 7.582125965096081e-05
- warmup_ratio: 0.04
- min_lr: 5.001256479531022e-11
- factor: 0.94
Training arguments:
output_dir: /mimer/NOBACKUP/groups/naiss2023-6-290/stefano/models//PROTAC-Splitter-EncoderDecoder-lr_reduce-opt25-rand-smiles
learning_rate: 5e-05
optim: adamw_torch
lr_scheduler_type: reduce_lr_on_plateau
lr_scheduler_kwargs: {'min_lr': 1e-06}
adam_beta1: 0.9
adam_beta2: 0.999
adam_epsilon: 1e-08
predict_with_generate: True
generation_config: GenerationConfig { "max_length": 512, "max_new_tokens": 512 }
generation_max_length: 512
batch_eval_metrics: False
group_by_length: True
per_device_train_batch_size: 128
per_device_eval_batch_size: 64
gradient_accumulation_steps: 1
auto_find_batch_size: True
fp16: True
fp16_full_eval: True
half_precision_backend: auto
use_cpu: False
dataloader_num_workers: 8
dataloader_prefetch_factor: None
max_steps: 10000
num_train_epochs: -1
save_steps: 5000
save_strategy: steps
eval_steps: 2500
eval_delay: 5000
eval_strategy: steps
save_total_limit: 2
load_best_model_at_end: True
metric_for_best_model: all_ligands_equal
include_inputs_for_metrics: True
eval_on_start: False
log_level: debug
logging_steps: 500
disable_tqdm: True
report_to: ['tensorboard']
save_only_model: False
push_to_hub: True
push_to_hub_model_id: PROTAC-Splitter-EncoderDecoder-lr_reduce-opt25-rand-smiles
push_to_hub_organization: ailab-bio
hub_model_id: ailab-bio/PROTAC-Splitter-EncoderDecoder-lr_reduce-opt25-rand-smiles
hub_strategy: checkpoint
hub_private_repo: True
seed: 42
data_seed: 42
warmup_steps: 800