speech_recognition/FunASR
1
0
Fork 0
mirror of https://github.com/modelscope/FunASR synced 2025-09-15 14:48:36 +08:00
Code Issues Actions Packages Projects Releases Wiki Activity

auto frontend

Browse Source
This commit is contained in:
游雁 2024-06-05 16:54:33 +08:00
parent 545d69ae92
commit a6441441cb
2 changed files with 4 additions and 4 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

5
funasr/models/llm_asr/adaptor.py
View File

@ -1,5 +1,7 @@
import torch import torch
import torch.nn as nn import torch.nn as nn
import torch.nn.functional as F
from funasr.models.transformer.utils.nets_utils import make_pad_mask
from funasr.register import tables from funasr.register import tables
@ -119,9 +121,8 @@ class Transformer(nn.Module):
x = self.linear2(x) x = self.linear2(x)
olens = None olens = None
if ilens is not None:
olens = (ilens - 1) // self.k + 1 olens = (ilens - 1) // self.k + 1
mask = (~make_pad_mask(olens)[:, None, :]).to(x.device) masks = (~make_pad_mask(olens)[:, None, :]).to(x.device)
for layer, block in enumerate(self.blocks): for layer, block in enumerate(self.blocks):
x, masks = block(x, masks) x, masks = block(x, masks)
return x, olens return x, olens

1
funasr/train_utils/trainer_ds.py
View File

@ -621,7 +621,6 @@ class Trainer:
self.train_acc_avg = train_acc_avg.detach().cpu().item() / self.world_size self.train_acc_avg = train_acc_avg.detach().cpu().item() / self.world_size
def forward_step(self, model, batch, loss_dict={}): def forward_step(self, model, batch, loss_dict={}):
dtype = torch.bfloat16
with maybe_autocast(dtype=self.dtype, use_deepspeed=self.use_deepspeed): with maybe_autocast(dtype=self.dtype, use_deepspeed=self.use_deepspeed):
retval = model(**batch) retval = model(**batch)