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 2024-06-07 11:40:46 +08:00
parent c350b6becb
commit 8d7f76af46
1 changed files with 6 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

7
funasr/models/llm_asr/model.py
View File

@ -480,10 +480,15 @@ class LLMASR2(nn.Module):
inputs_embeds = self.llm.model.get_input_embeddings()(input_ids)
batch_size, token_num, dims = inputs_embeds.shape
_, l, _ = encoder_out.shape
fbank_mask[fbank_mask < 0] = 0
fbank_fake_lens = fbank_mask.sum(-1)
# _, l, _ = encoder_out.shape
for batch_idx in range(batch_size):
l = fbank_fake_lens[batch_idx].item()
fbank_beg_idx = fbank_beg[batch_idx, 0].item()
min_len = min(l, inputs_embeds.shape[1] - fbank_beg_idx)
inputs_embeds[batch_idx, fbank_beg_idx : fbank_beg_idx + min_len, :] = encoder_out[
batch_idx, :min_len, :
]