Skip to content

fix: BiLSTM fp32 (bf16 NaN collapse) #132

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
ronaldtse merged 1 commit into from
Oct 9, 2026
Merged

fix: BiLSTM fp32 (bf16 NaN collapse) #132

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
10 changes: 6 additions & 4 deletions train_bilstm_hf.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -159,9 +159,12 @@ def forward(self, ids):
for _ep in range(EPOCHS):
for ids, lab in loader:
ids, lab = ids.cuda(), lab.cuda()
with torch.autocast("cuda", torch.bfloat16):
logits = model(ids)
logits = model(ids) # fp32: LSTMs NaN-collapse under bf16 autocast
loss = ce(logits.reshape(-1, n_classes), lab.reshape(-1))
if not torch.isfinite(loss):
print(f"[nan] step {step}: non-finite loss, skipping batch", flush=True)
opt.zero_grad()
continue
loss.backward()
torch.nn.utils.clip_grad_norm_(model.parameters(), 1.0)
opt.step()
Expand All @@ -186,8 +189,7 @@ def tag(text: str) -> str:
skel, _ = HP.split_planes(text)
ids = [vocab.get(ch, 1) for ch in skel] or [0]
t = torch.tensor([ids], device="cuda")
with torch.autocast("cuda", torch.bfloat16):
logits = model(t)
logits = model(t)
preds = logits[0].argmax(-1).tolist()
return "".join(ch + (classes[preds[i]] if i < len(preds) else "")
for i, ch in enumerate(skel))
Expand Down
Loading