Skip to content

fix(train): size rotary cache to config.sequence_len, not 10x #557

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

fix(train): size rotary cache to config.sequence_len, not 10x #557

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
8 changes: 6 additions & 2 deletions train.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -140,8 +140,12 @@ def __init__(self, config):
str(i): nn.Embedding(config.vocab_size, kv_dim)
for i in range(config.n_layer) if has_ve(i, config.n_layer)
})
# Rotary embeddings
self.rotary_seq_len = config.sequence_len * 10
# Rotary embeddings. Sized to the configured sequence length — forward
# asserts T <= cos.size(1), so any T <= config.sequence_len fits. The
# previous `* 10` overshoot allocated 10x more cos/sin entries than
# could ever be indexed (forward never sees T > config.sequence_len),
# wasting ~90% of the buffer.
self.rotary_seq_len = config.sequence_len
cos, sin = self._precompute_rotary_embeddings(self.rotary_seq_len, head_dim)
self.register_buffer("cos", cos, persistent=False)
self.register_buffer("sin", sin, persistent=False)
Expand Down