Skip to content

Tied word embeddings #1260

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.

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

Merged
merged 1 commit into from
Oct 9, 2024
Merged
Changes from all commits
Commits
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
10 changes: 10 additions & 0 deletions torchchat/model.py
Original file line number Diff line number Diff line change
Expand Up @@ -281,6 +281,8 @@ class TransformerArgs:
# Optional biases
attention_bias: bool = False
feed_forward_bias: bool = False
# Whether or not to tie the input word embeddings to the output
tie_word_embeddings: bool = False

def __post_init__(self):
if self.n_local_heads == -1:
Expand Down Expand Up @@ -632,12 +634,20 @@ def __init__(self, config: TransformerArgs) -> None:
if config.stage_idx == config.n_stages - 1:
self.norm = RMSNorm(config.dim, eps=config.norm_eps)
self.output = nn.Linear(config.dim, config.vocab_size, bias=False)
if config.tie_word_embeddings:
self.output.weight = self.tok_embeddings.weight
else:
self.norm = None
self.output = None

self.max_batch_size = -1
self.max_seq_length = -1
self._register_load_state_dict_pre_hook(self.load_hook)

def load_hook(self, state_dict, prefix, *args):
"""Handle tied embeddings at load time"""
if self.config.tie_word_embeddings:
state_dict.setdefault("model.output.weight", state_dict["model.tok_embeddings.weight"])

def setup_caches(self, max_batch_size, max_seq_length, cache_lanes: int = 1):
if (
Expand Down
Loading