From 8d7f76af46cf0e77317ec8e84fcce6f208f24204 Mon Sep 17 00:00:00 2001
From: 游雁 <zhifu.gzf@alibaba-inc.com>
Date: 星期五, 07 六月 2024 11:40:46 +0800
Subject: [PATCH] auto frontend
---
funasr/models/llm_asr/model.py | 35 +++++++++++++++++++++--------------
1 files changed, 21 insertions(+), 14 deletions(-)
diff --git a/funasr/models/llm_asr/model.py b/funasr/models/llm_asr/model.py
index 411b59d..b139123 100644
--- a/funasr/models/llm_asr/model.py
+++ b/funasr/models/llm_asr/model.py
@@ -468,7 +468,7 @@
if len(speech_lengths.size()) > 1:
speech_lengths = speech_lengths[:, 0]
- batch_size = speech.shape[0]
+ batch_size, frames, _ = speech.shape
# audio encoder
encoder_out, encoder_out_lens = self.audio_encoder(speech.permute(0, 2, 1), speech_lengths)
@@ -476,23 +476,23 @@
# audio_adaptor
encoder_out, encoder_out_lens = self.audio_adaptor(encoder_out, encoder_out_lens)
- input_ids[input_ids == -1] = 0
- input_ids[input_ids == -100] = 0
- if hasattr(self.llm.model, "embed_tokens"):
- inputs_embeds = self.llm.model.embed_tokens(input_ids)
- elif hasattr(self.llm.model.model, "embed_tokens"):
- inputs_embeds = self.llm.model.model.embed_tokens(input_ids)
- else:
- inputs_embeds = self.llm.model.model.model.embed_tokens(input_ids)
+ input_ids[input_ids < 0] = 0
+ 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):
- fbank_beg_idx = fbank_beg[batch_idx, 0].item()
- inputs_embeds[batch_idx, fbank_beg_idx : fbank_beg_idx + l, :] = encoder_out[
- batch_idx, :l, :
- ]
+ 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, :
+ ]
+ labels_ids[labels_ids == -1] = -100
model_outputs = self.llm(
inputs_embeds=inputs_embeds, attention_mask=attention_mask, labels=labels_ids
)
@@ -505,6 +505,13 @@
stats["acc"] = acc_att
stats["loss"] = torch.clone(loss.detach())
+ stats["batch_size"] = batch_size
+ stats["batch_size_x_frames"] = frames * batch_size
+ stats["batch_size_real_frames"] = speech_lengths.sum().item()
+ stats["padding_frames"] = stats["batch_size_x_frames"] - stats["batch_size_real_frames"]
+ stats["batch_size_x_tokens"] = token_num * batch_size
+ stats["batch_size_real_tokens"] = attention_mask.sum().item()
+ stats["padding_tokens"] = stats["batch_size_x_tokens"] - stats["batch_size_real_tokens"]
# force_gatherable: to-device and to-tensor if scalar for DataParallel
if self.length_normalized_loss:
--
Gitblit v1.9.1