Skip to content

fix(data): harden process_token_dict_to_mappings with explicit dtypes, O(1) lookup, and assertions #304

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
eppaneamd wants to merge 2 commits into
base: main
Choose a base branch
from
Open

fix(data): harden process_token_dict_to_mappings with explicit dtypes, O(1) lookup, and assertions #304

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
dc60d62
fix(data): explicit dtypes, O(1) set lookup, and assertions in proces…
eppaneamd May 11, 2026
8ade0e8
style: apply black formatting
eppaneamd May 12, 2026
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
19 changes: 15 additions & 4 deletions angelslim/compressor/speculative/train/data/data_utils.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -133,10 +133,21 @@ def process_token_dict_to_mappings(
used_tokens = [key for key, freq in top_N]
used_tokens.sort()

d2t = [used_tokens[i] - i for i in range(len(used_tokens))]
t2d = [i in used_tokens for i in range(target_vocab_size)]
d2t = torch.tensor(d2t)
t2d = torch.tensor(t2d)
used_set = set(used_tokens)
d2t = torch.tensor(
[used_tokens[i] - i for i in range(len(used_tokens))],
dtype=torch.int64, # must match register_buffer dtype in Eagle3LlamaForCausalLM
)
t2d = torch.tensor(
[i in used_set for i in range(target_vocab_size)],
dtype=torch.bool, # must match register_buffer dtype in Eagle3LlamaForCausalLM
)

assert d2t.shape == (draft_vocab_size,), f"d2t shape {d2t.shape} != ({draft_vocab_size},)"
assert t2d.shape == (target_vocab_size,), f"t2d shape {t2d.shape} != ({target_vocab_size},)"
assert (
t2d.sum().item() == draft_vocab_size
), f"t2d has {t2d.sum().item()} True entries, expected {draft_vocab_size}"

return d2t, t2d

Expand Down
Loading