mrq/vall-e
5
9
Fork 0
Code Issues 8 Pull Requests Packages Projects Releases Wiki Activity

do not like that

Browse Source
This commit is contained in:
mrq 2025-02-27 23:59:56 -06:00
parent f4f435d7f5
commit 93feb5660f
2 changed files with 11 additions and 20 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

3
vall_e/models/ar_nar_v2.py
View File

@ -1037,7 +1037,8 @@ def example_usage():
texts, proms, resps, tasks = sample_data()
stats = {"step": i}
stats |= engine.traverse(phns_list=texts, proms_list=proms, resps_list=resps, task_list=tasks, training=True)
with torch.autograd.set_detect_anomaly(cfg.trainer.detect_grad_anomaly):
stats |= engine.traverse(phns_list=texts, proms_list=proms, resps_list=resps, task_list=tasks, training=True)
stats |= {"grad_norm": engine.get_global_grad_norm()}
tqdm.write(f"{stats}")

28
vall_e/models/base_v2.py
View File

@ -939,18 +939,13 @@ class Base_V2(nn.Module):
name = f'{name}[{level}]'
sequence = token if token.dim() <= 1 else token[:, level]
nll, metrics = _calc_loss( logits[batch_index][level][start:end], sequence.long(), causal, level )
nll, metrics = _calc_loss( logits[batch_index][level][start:end], sequence.long(), causal )
else:
sequence = token.t()
nll, metrics = _calc_loss( logits[batch_index][:, start:end], sequence.long(), causal )
for level in enumerate(self.n_resp_levels):
loss_key = f'{name}[{level}].nll'
if loss_key not in loss:
loss[loss_key] = []
loss[loss_key].append( nll[level] * loss_factor )
nll = None
if nll is not None:
nll = nll.sum()
loss_key = f'{name}.nll'
acc_key = f'{name}.acc'
@ -982,7 +977,7 @@ class Base_V2(nn.Module):
sequence = [ x if x.dim() <= 1 else x[:, level] for x in target ]
sequence = _join( sequence, torch.tensor(self.ignore_index, device=sequence[-1].device) )
nll, metrics = _calc_loss( logits[batch_index][level], sequence.long(), causal, level )
nll, metrics = _calc_loss( logits[batch_index][level], sequence.long(), causal )
else:
nlls = []
accs = []
@ -991,16 +986,11 @@ class Base_V2(nn.Module):
sequence = [ x if x.dim() <= 1 else x[:, level] for x in target ]
sequence = _join( sequence, torch.tensor(self.ignore_index, device=sequence[-1].device) )
nll, metrics = _calc_loss( logit, sequence, causal, level )
if nll:
nlls.append( nll )
if metrics:
accs.append( metrics )
if nlls:
nll = sum(nlls) / len(nlls)
if accs:
accs = sum(accs) / len(accs)
nlls.append( nll )
nll = sum(nlls)
accs = sum(accs) / len(accs)
if nll is not None:
if 'nll' not in loss: