Skip to content

Fix weight decay #6

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
Niccolo-Ajroldi wants to merge 2 commits into
base: main
Choose a base branch
from
Open

Fix weight decay #6

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
5ab076a
no wd on bias and mamba A,D params
Niccolo-Ajroldi Dec 5, 2024
2359273
add names to normaliz layers, allows to exclude them from wd params
Niccolo-Ajroldi Dec 5, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
14 changes: 11 additions & 3 deletions mad/model/language_model.py
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,3 +1,4 @@
from collections import OrderedDict
import torch
import typing as tp
from torch import nn
Expand Down Expand Up @@ -46,9 +47,16 @@ def __init__(self,

self.model = nn.ModuleList([])
for layer, layer_cfg in zip(layers, layer_cfgs):
self.model.append(nn.Sequential(norm(layer_cfg['dim']), layer(**layer_cfg)))

self.unembed = nn.Sequential(norm(layer_cfg['dim']), nn.Linear(dim, vocab_size))
self.model.append(nn.Sequential(OrderedDict([
('norm', norm(layer_cfg['dim'])),
('layer', layer(**layer_cfg))
])))

self.unembed = nn.Sequential(OrderedDict([
('norm', norm(layer_cfg['dim'])),
('lm_head', nn.Linear(dim, vocab_size))
]))

self.apply(self._init_weights)

def embed(self,
Expand Down
23 changes: 17 additions & 6 deletions mad/model/pl_model_wrapper.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -93,18 +93,29 @@ def test_step(self,
return self.phase_step(batch, batch_idx, phase='test')

def configure_optimizers(self) -> tp.Union[torch.optim.Optimizer, tp.Dict[str, tp.Any]]:
# param groups
decay_params, no_decay_params = [], []
for n, p in self.model.named_parameters():
if p.requires_grad:
if not getattr(p, '_no_weight_decay', False) and ("bias" not in n) and ("norm" not in n):
decay_params.append(p)
else:
no_decay_params.append(p)
param_groups = [
{"params": decay_params, "weight_decay": self.mad_config.weight_decay},
{"params": no_decay_params, "weight_decay": 0.0},
]

# optimizer:
if self.mad_config.optimizer == 'adamw':
optimizer = torch.optim.AdamW(
self.parameters(),
lr=self.mad_config.lr,
weight_decay=self.mad_config.weight_decay
param_groups,
lr=self.mad_config.lr
)
elif self.mad_config.optimizer == 'sgd':
optimizer = torch.optim.SGD(
self.parameters(),
lr=self.mad_config.lr,
weight_decay=self.mad_config.weight_decay
param_groups,
lr=self.mad_config.lr
)
else:
raise ValueError(f"invalid optimizer: {self.mad_config.optimizer}")
Expand Down