Skip to content

Fix masked softmax in inference #9

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
mcognetta wants to merge 3 commits into
base: main
Choose a base branch
from
Open

Fix masked softmax in inference #9

Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
de43968
fix masked softmax
mcognetta Oct 6, 2025
c3b54d9
fix in inference_each also
mcognetta Nov 16, 2025
4017653
add guard for degenerate case where no legal moves are available
mcognetta Feb 22, 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 maia2/inference.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -60,8 +60,13 @@ def get_preds(model, dataloader, all_moves_dict_reversed):
legal_moves = legal_moves.to(device)

logits_maia, _, logits_value = model(boards, elos_self, elos_oppo)
logits_maia_legal = logits_maia * legal_moves
probs = logits_maia_legal.softmax(dim=-1).cpu().tolist()
if not legal_moves.any(dim=-1).all():
# no legal moves available, so just output zeros for all probabilities
# this should prevent any moves from being present in the output dictionary
probs = torch.zeros_like(logits_maia, device='cpu').tolist()
else:
logits_maia_legal = logits_maia + legal_moves.log()
probs = logits_maia_legal.softmax(dim=-1).cpu().tolist()

logits_value = (logits_value / 2 + 0.5).clamp(0, 1).cpu().tolist()

Expand Down Expand Up @@ -154,8 +159,14 @@ def inference_each(model, prepared, fen, elo_self, elo_oppo):
legal_moves = legal_moves.unsqueeze(dim=0).to(device)

logits_maia, _, logits_value = model(board_input, elo_self, elo_oppo)
logits_maia_legal = logits_maia * legal_moves
probs = logits_maia_legal.softmax(dim=-1).cpu().tolist()

if not legal_moves.any(dim=-1).all():
# no legal moves available, so just output zeros for all probabilities
# this should prevent any moves from being present in the output dictionary
probs = torch.zeros_like(logits_maia, device='cpu').tolist()
else:
logits_maia_legal = logits_maia + legal_moves.log()
probs = logits_maia_legal.softmax(dim=-1).cpu().tolist()

logits_value = (logits_value / 2 + 0.5).clamp(0, 1).item()

Expand Down