Skip to content

[BIT-588] Create topk_tensor on origin device #926

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

Merged
opentaco merged 2 commits into from
Sep 23, 2022
Merged

[BIT-588] Create topk_tensor on origin device #926

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
58adeae
Create topk_tensor on origin device
opentaco Sep 23, 2022
9b5190f
Merge branch 'nobunaga' into feature/BIT-588/device_unravel_topk_toke…
opentaco Sep 23, 2022
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
3 changes: 2 additions & 1 deletion bittensor/utils/tokenizer_utils.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -878,7 +878,8 @@ def unravel_topk_token_phrases(compact_topk: torch.Tensor, topk: int, ignore_ind
max_len = phrase_len.max() # determine width of topk_tensor as max len of all phrase lists (with prob in front)

# Initialize topk_tensor with ignore_index + 2, since decrement with 2 follows to remove token offset later
topk_tensor = (ignore_index + 2) * torch.ones((batch_size * (topk + 1), max_len)) # [batch_size * (topk + 1), max_len]
topk_tensor = torch.ones((batch_size * (topk + 1), max_len), device=compact_topk.device)
topk_tensor *= ignore_index + 2 # [batch_size * (topk + 1), max_len]

# Insert phrases of each unique length as block into topk_tensor
for unique_len in phrase_len.unique():
Expand Down