Upload best_hyperparameters.md with huggingface_hub
Browse files- best_hyperparameters.md +63 -0
best_hyperparameters.md
ADDED
|
@@ -0,0 +1,63 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
Number of Optuna trials: 25
|
| 2 |
+
|
| 3 |
+
Best trial objective: 0.5828 (best trial number: 18)
|
| 4 |
+
|
| 5 |
+
Best hyperparameters:
|
| 6 |
+
- learning_rate: 0.0002004749428706299
|
| 7 |
+
- warmup_ratio: 0.01
|
| 8 |
+
- num_cycles: 2.5
|
| 9 |
+
|
| 10 |
+
Training arguments:
|
| 11 |
+
- output_dir: /mimer/NOBACKUP/groups/naiss2023-6-290/stefano/models//PROTAC-Splitter-EncoderDecoder-lr_cosine-opt25
|
| 12 |
+
- learning_rate: 5e-05
|
| 13 |
+
- optim: adamw_torch
|
| 14 |
+
- lr_scheduler_type: cosine
|
| 15 |
+
- lr_scheduler_kwargs: {}
|
| 16 |
+
- adam_beta1: 0.9
|
| 17 |
+
- adam_beta2: 0.999
|
| 18 |
+
- adam_epsilon: 1e-08
|
| 19 |
+
- predict_with_generate: True
|
| 20 |
+
- generation_config: GenerationConfig {
|
| 21 |
+
"max_length": 512,
|
| 22 |
+
"max_new_tokens": 512
|
| 23 |
+
}
|
| 24 |
+
|
| 25 |
+
- generation_max_length: 512
|
| 26 |
+
- batch_eval_metrics: False
|
| 27 |
+
- group_by_length: True
|
| 28 |
+
- per_device_train_batch_size: 128
|
| 29 |
+
- per_device_eval_batch_size: 64
|
| 30 |
+
- gradient_accumulation_steps: 1
|
| 31 |
+
- auto_find_batch_size: True
|
| 32 |
+
- fp16: True
|
| 33 |
+
- fp16_full_eval: True
|
| 34 |
+
- half_precision_backend: auto
|
| 35 |
+
- use_cpu: False
|
| 36 |
+
- dataloader_num_workers: 8
|
| 37 |
+
- dataloader_prefetch_factor: None
|
| 38 |
+
- max_steps: 10000
|
| 39 |
+
- num_train_epochs: -1
|
| 40 |
+
- save_steps: 5000
|
| 41 |
+
- save_strategy: steps
|
| 42 |
+
- eval_steps: 2500
|
| 43 |
+
- eval_delay: 5000
|
| 44 |
+
- eval_strategy: steps
|
| 45 |
+
- save_total_limit: 2
|
| 46 |
+
- load_best_model_at_end: True
|
| 47 |
+
- metric_for_best_model: all_ligands_equal
|
| 48 |
+
- include_inputs_for_metrics: True
|
| 49 |
+
- eval_on_start: False
|
| 50 |
+
- log_level: debug
|
| 51 |
+
- logging_steps: 500
|
| 52 |
+
- disable_tqdm: True
|
| 53 |
+
- report_to: ['tensorboard']
|
| 54 |
+
- save_only_model: False
|
| 55 |
+
- push_to_hub: True
|
| 56 |
+
- push_to_hub_model_id: PROTAC-Splitter-EncoderDecoder-lr_cosine-opt25
|
| 57 |
+
- push_to_hub_organization: ailab-bio
|
| 58 |
+
- hub_model_id: ailab-bio/PROTAC-Splitter-EncoderDecoder-lr_cosine-opt25
|
| 59 |
+
- hub_strategy: checkpoint
|
| 60 |
+
- hub_private_repo: True
|
| 61 |
+
- seed: 42
|
| 62 |
+
- data_seed: 42
|
| 63 |
+
- warmup_steps: 699
|