commit from alexey
Browse files- config.json +55 -0
config.json
ADDED
@@ -0,0 +1,55 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"model_path": "/mnt/10tb/home/shmelev/dnalm/downstream_tasks/APARENT/bert_base_sparse_rope_4096_bs256_lr_5e-05_wd0.01_fp16_from_425k/",
|
3 |
+
"log_interval": 250,
|
4 |
+
"valid_interval": 1000,
|
5 |
+
"save_interval": null,
|
6 |
+
"save_best": true,
|
7 |
+
"use_generate_on_valid": false,
|
8 |
+
"init_checkpoint": "/mnt/10tb/home/shmelev/dnalm/downstream_tasks/APARENT/bert_base_sparse_rope_4096_bs256_lr_5e-05_wd0.01_fp16_from_425k/model_best_from_s3.pth",
|
9 |
+
"skip_used_data": false,
|
10 |
+
"reset_lr": true,
|
11 |
+
"reset_iteration": true,
|
12 |
+
"reset_optimizer": true,
|
13 |
+
"lr": 5e-05,
|
14 |
+
"batch_size": 32,
|
15 |
+
"iters": 500000,
|
16 |
+
"gradient_accumulation_steps": 1,
|
17 |
+
"fp16": true,
|
18 |
+
"fp16_allreduce": false,
|
19 |
+
"apex_opt_lvl": "O2",
|
20 |
+
"min_loss_scale": null,
|
21 |
+
"max_loss_scale": 16777216,
|
22 |
+
"clip_grad_norm": null,
|
23 |
+
"clip_grad_value": null,
|
24 |
+
"early_stopping_patience": null,
|
25 |
+
"lr_scheduler": "constant_with_warmup",
|
26 |
+
"num_warmup_steps": 1500,
|
27 |
+
"num_training_steps": null,
|
28 |
+
"use_lr_drop": false,
|
29 |
+
"lr_drop_factor": 0.1,
|
30 |
+
"lr_drop_patience": 10,
|
31 |
+
"lr_drop_threshold": 0.0001,
|
32 |
+
"lr_drop_threshold_mode": "rel",
|
33 |
+
"lr_drop_cooldown": 0,
|
34 |
+
"lr_drop_min_lr": 0.0,
|
35 |
+
"lr_drop_eps": 1e-08,
|
36 |
+
"optimize_metric": "pearsonr2",
|
37 |
+
"optimize_mode": "max",
|
38 |
+
"train_csv": "/mnt/10tb/home/shmelev/dnalm/downstream_tasks/APARENT/dataset_itself/APARENT_train.csv",
|
39 |
+
"test_csv": "/mnt/10tb/home/shmelev/dnalm/downstream_tasks/APARENT/dataset_itself/APARENT_test.csv",
|
40 |
+
"seed": 42,
|
41 |
+
"input_seq_len": 256,
|
42 |
+
"data_n_workers": 2,
|
43 |
+
"model_cfg": "/mnt/10tb/home/shmelev/dnalm/data/configs/L12-H768-A12-V32k-L4096-preln-sparse-rope.json",
|
44 |
+
"model_cls": "src.gena_lm.modeling_bert:BertForAPARENTSequenceRegression",
|
45 |
+
"tokenizer": "/mnt/10tb/home/shmelev/dnalm/data/tokenizers/human/BPE_32k/",
|
46 |
+
"optimizer": "AdamW",
|
47 |
+
"weight_decay": 0.0001,
|
48 |
+
"ENV": {
|
49 |
+
"CUDA_VISIBLE_DEVICES": "0"
|
50 |
+
},
|
51 |
+
"HVD_INIT": true,
|
52 |
+
"HVD_SIZE": 1,
|
53 |
+
"MACHINE": "bio-protein",
|
54 |
+
"COMMIT": "2513d0c4ee891e7bcd224b9c65e88e27e0fd1892"
|
55 |
+
}
|