Skip to content

fix: guard division-by-zero and uninitialized state in metrics/masking code #1576

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
mooreneural wants to merge 4 commits into
base: main
Choose a base branch
from
+9 −4
Open

fix: guard division-by-zero and uninitialized state in metrics/masking code #1576

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
0201acc
fix: guard against division-by-zero and uninitialized state in metric…
mooreneural May 18, 2026
9184a26
Merge branch 'main' into mooreneural/BioNeMoore
mooreneural May 19, 2026
0ad7a85
Merge branch 'main' into mooreneural/BioNeMoore
mooreneural May 26, 2026
7c89e10
Merge branch 'main' into mooreneural/BioNeMoore
mooreneural May 27, 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
1 change: 1 addition & 0 deletions bionemo-recipes/interpretability/sparse_autoencoders/sae/src/sae/eval/dead_latents.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -67,6 +67,7 @@ def __init__(

# Tracks activity since last reset
self.recently_active = torch.zeros(hidden_dim, dtype=torch.bool, device=device)
self._last_avg_nonzero: float = 0.0

@torch.no_grad()
def update(self, codes: torch.Tensor) -> None:
Expand Down
11 changes: 8 additions & 3 deletions bionemo-recipes/recipes/codonfm_ptl_te/src/data/preprocess/mlm_memmap.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -66,7 +66,9 @@ def process_item( # noqa: D417
prob_input_sequence = np.ones_like(input_sequence_toks) * mlm_probability
if codon_weights is not None:
pos_weight = codon_weights[input_sequence_toks]
pos_weight[1:-1] = pos_weight[1:-1] / pos_weight[1:-1].mean()
mean_weight = pos_weight[1:-1].mean()
if mean_weight > 0:
pos_weight[1:-1] = pos_weight[1:-1] / mean_weight
prob_input_sequence = prob_input_sequence * pos_weight
prob_input_sequence = np.clip(prob_input_sequence, 0.05, 0.4)
mask_indices = np.random.binomial(1, prob_input_sequence).astype(bool) # noqa: NPY002
Expand All @@ -82,10 +84,13 @@ def process_item( # noqa: D417
masked_input_sequence_toks[indices_replaced] = tokenizer.mask_token_id

if random_replace_prob > 0.0:
remaining_prob = 1.0 - mask_replace_prob
# Conditional probability of random replacement given the token was not replaced by [MASK].
# Guard against mask_replace_prob == 1.0 (no remaining positions) or rounding above 1.0.
conditional_prob = min(random_replace_prob / remaining_prob, 1.0) if remaining_prob > 0.0 else 0.0
indices_random = np.random.binomial( # noqa: NPY002
1,
(np.ones_like(mask_indices).astype(bool) & mask_indices & ~indices_replaced)
* (random_replace_prob / (1 - mask_replace_prob)),
(np.ones_like(mask_indices).astype(bool) & mask_indices & ~indices_replaced) * conditional_prob,
).astype(bool)
valid_tokens = np.setdiff1d(
np.arange(tokenizer.vocab_size),
Expand Down
1 change: 0 additions & 1 deletion bionemo-recipes/recipes/evo2_megatron/src/bionemo/evo2/data/megatron/hyena/evo2_dataset.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -38,7 +38,6 @@ class Evo2Dataset(GPTDataset):
MAX_TAG_LEN = 2048

VALID_DNA_AND_DEGENERATE: ClassVar[set[int]] = {
45,
45,
65,
66,
Expand Down
Loading