File size: 1,325 Bytes
e09b20c
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
from train import train

params = [
    {
    "testing": False,
    "seed": 97,
    "data_dir": f"../datasets/exhaustive_{nb_jobs}_{nb_machines}/top_{top_k}",
    "n_embd": n_embd,
    "n_head": n_head,
    "n_layer": n_layer,
    "ff_width": 4,
    "intermediate_schedules": True,
    "train_batch_size": 512,
    "val_batch_size": 256,
    "nb_epochs": 5,
    "early_stopping_patience": 15,
    "dropout": 0.0,
    "checkpoint_interval_ratio": 1.0,
    "decay_lr": True,
    "lr_partitions_ratios": [0.66],
    "init_lr": 1e-4,
    "max_lr": 1e-3,
    "min_lr": 5e-5,
    "lr_warmup_iters_ratio": 0.1,
    "lr_decay_iters_ratio": 0.95,
    "beta1": 0.9,
    "beta2": 0.95,
    "weight_decay": wd,
    "grad_clip": 1.0,
    "compile": "",
    "compile_mode": "default",
    "save_only_last_checkpoint": True,
    "output_dir": f"../datasets/exhaustive_{nb_jobs}_{nb_machines}/top_{top_k}/train_{model_size_code}_Wd{wd_str}",
    }
    for nb_jobs in [7, 8, 9]
    for nb_machines in [2, 3, 4, 5, 6]
    for top_k in [0, 1, 2, 3, 4]
    for model_size_code, n_embd, n_head, n_layer in [("Bm", 256, 16, 8), ("Mm", 128, 8, 4), ("Sm", 64, 4, 2)] # Bm: Big model, Mm: Medium model, Sm: Small model
    for wd, wd_str in [(1e-1, "1e-1"), (1e-0, "1e-0"), (5e-0, "5e-0"), (10e-0, "10e-0")]
]

for param in params:
    train(**param)