Skip to content

FIX TypeError: normal_() got an unexpected keyword argument 'generator' #723 #791

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
wants to merge 1 commit into
base: master
Choose a base branch
from
Open

FIX TypeError: normal_() got an unexpected keyword argument 'generator' #723 #791

Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
5 changes: 3 additions & 2 deletions train_gpt2.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -146,18 +146,19 @@ def __init__(self, config):
self.init_rng.manual_seed(42)
self.apply(self._init_weights)

@torch.no_grad()
def _init_weights(self, module):
if isinstance(module, nn.Linear):
# apply special scaled init to the residual projections, per GPT-2 paper
std = 0.02 if not hasattr(module, 'LLMC_RESIDUAL_SCALE_FLAG') else 0.02/math.sqrt(2 * self.config.n_layer)
# we want to skip initializing lm_head, which shares parameters with wte
# and wte was already initialized down below during the Embedding init
if not hasattr(module, 'LLMC_SKIP_INIT'):
torch.nn.init.normal_(module.weight, mean=0.0, std=std, generator=self.init_rng)
module.weight.normal_(mean=0.0, std=std, generator=self.init_rng)
if module.bias is not None:
torch.nn.init.zeros_(module.bias)
elif isinstance(module, nn.Embedding):
torch.nn.init.normal_(module.weight, mean=0.0, std=0.02, generator=self.init_rng)
module.weight.normal_(mean=0.0, std=0.02, generator=self.init_rng)

def forward(self, idx, targets=None, return_logits=True):
device = idx.device
Expand Down