Functions — github.com/ASLP-lab/OSUM-Pangu

↓ 2 callersMethodsubsampling_rate

Export interface for c++ call, return subsampling_rate of the model

wenet/transformer/asr_model.py:389

↓ 2 callersFunctionsubsequent_chunk_mask

Create mask for subsequent steps (size, size) with chunk size, this is for streaming encoder Args: size (int): size of mask

wenet/utils/mask.py:88

↓ 2 callersMethodtar_file_and_group

(self, sample)

tools/extract_shard_data.py:70

↓ 2 callersMethodtie_or_clone_weights

Tie or clone module weights (between word_emb and output_layer) depending of whether we are using TorchScript or not

wenet/transformer/decoder.py:283

↓ 2 callersFunctiontokenize_by_bpe_model

(sp, txt)

wenet/text/tokenize_utils.py:18

↓ 2 callersMethodtotal_score

(self)

wenet/transformer/search.py:92

↓ 2 callersMethodupdate_context

(self, context_graph, prefix_score, word_id)

wenet/transformer/search.py:99

↓ 2 callersFunctionwidth

(string)

tools/compute-cer.py:250

↓ 2 callersFunctionwidth

(string)

tools/compute-wer.py:243

↓ 1 callersFunctionBigDataset

Construct dataset from arguments We have two shuffle stage in the Dataset. The first is global shuffle at shards tar/raw file level.

wenet/dataset/dataset.py:330

↓ 1 callersMethodCTL

(self, x, y, negs, mask=None)

wenet/ctl_model/asr_model_ctl.py:260

↓ 1 callersFunctionDataset

Construct dataset from arguments for ssl model We have two shuffle stage in the Dataset. The first is global shuffle at shards tar/r

wenet/ssl/init_dataset.py:42

↓ 1 callersMethod__init__

( self, size: int, linear_units: int, kernel_size: int, dropout_rate:

wenet/branchformer/cgmlp.py:137

↓ 1 callersMethod__init__

( self, input_size: int, output_size: int = 256, use_attn: bool = True,

wenet/branchformer/encoder.py:32

↓ 1 callersMethod__init__

( self, idim, l_order, r_order, threshold=1.0, dropout=0.1,

wenet/paraformer/paraformer.py:39

↓ 1 callersMethod__init__

( self, idim, l_order, r_order, threshold=1.0, dropout=0.1,

wenet/paraformer/cif.py:26

↓ 1 callersMethod__init__

( self, vocab_size: int, encoder_output_size: int, attention_heads: int = 4,

wenet/transformer/decoder.py:63

↓ 1 callersMethod__init__

Construct CTC module Args: odim: dimension of outputs encoder_output_size: number of encoder projection units

wenet/transformer/ctc.py:73

↓ 1 callersMethod__init__

Construct an EncoderLayer object.

wenet/transformer/encoder_layer.py:150

↓ 1 callersMethod__init__

Construct DualConformerEncoder Support both the full context mode and the streaming mode separately

wenet/ctl_model/encoder.py:99

↓ 1 callersMethod__init__

(self, encoder, encoder_output_dim, llm_path,

wenet/llm_asr/llmasr_model_instruct_version.py:49

↓ 1 callersMethod_apply_mask

( self, xs: torch.Tensor, xs_masks: torch.Tensor)

wenet/ssl/w2vbert/w2vbert_model.py:253

↓ 1 callersMethod_apply_mask

( self, xs: torch.Tensor, xs_masks: torch.Tensor)

wenet/ssl/wav2vec2/wav2vec2_model.py:285

↓ 1 callersMethod_apply_mask_signal

( self, input: torch.Tensor, input_lens: torch.Tensor)

wenet/ssl/bestrq/bestrq_model.py:216

↓ 1 callersFunction_batch_beam_search

Perform batch beam search Args: logit: shape (batch_size, seq_length, vocab_size) masks: shape (batch_size, seq_leng

wenet/paraformer/search.py:200

↓ 1 callersMethod_cal_attn_score

( self, encoder_out: torch.Tensor, encoder_mask: torch.Tensor, hyps_pad: torch

wenet/transducer/transducer.py:187

↓ 1 callersMethod_cal_transducer_score

( self, encoder_out: torch.Tensor, encoder_mask: torch.Tensor, hyps_lens: torc

wenet/transducer/transducer.py:160

↓ 1 callersMethod_calc_att_loss

( self, encoder_out: torch.Tensor, encoder_mask: torch.Tensor, ys_pad: torch.T

wenet/paraformer/paraformer.py:218

↓ 1 callersMethod_calc_lfmmi_loss

(self, encoder_out, encoder_mask, text)

wenet/k2/model.py:100

↓ 1 callersMethod_compute_loss

(self, encoder_out: torch.Tensor, encoder_out_lens: torch.Tensor,

wenet/transducer/transducer.py:474

↓ 1 callersMethod_compute_loss

(self, input: torch.Tensor, target: torch.Tensor, mask: torch.Tensor)

wenet/ssl/bestrq/bestrq_model.py:279

↓ 1 callersMethod_compute_mlm_loss

(self, input: torch.Tensor, target: torch.Tensor, mask: torch.Tensor)

wenet/ssl/w2vbert/w2vbert_model.py:273

↓ 1 callersFunction_cosine_annealing

(initial_lr, step, max_steps, min_lr)

wenet/utils/scheduler.py:396

↓ 1 callersMethod_fill_fail_output

This function fills the fail arc for each trie node, it can be computed in linear time by performing a breadth-first search starting from the

wenet/utils/context_graph.py:176

↓ 1 callersFunction_filter_dict_fields

(input_dict, fields_to_keep)

wenet/paraformer/convert_paraformer_to_wenet_config_and_ckpt.py:46

↓ 1 callersMethod_forward

( self, x: torch.Tensor, mask: torch.Tensor, pos_emb: torch.Tensor, ma

wenet/branchformer/encoder_layer.py:111

↓ 1 callersMethod_forward

( self, x: torch.Tensor, mask: torch.Tensor, pos_emb: torch.Tensor, ma

wenet/e_branchformer/encoder_layer.py:93

↓ 1 callersMethod_forward_ctc

( self, encoder_out: torch.Tensor, encoder_mask: torch.Tensor, text: torch.Tensor,

wenet/k2/model.py:60

↓ 1 callersMethod_forward_encoder_blocks

( self, xs: torch.Tensor, xs_masks: torch.Tensor, pos_emb: torch.Tensor, mask_pad: torch.Tenso

wenet/ssl/w2vbert/w2vbert_model.py:296

↓ 1 callersMethod_forward_encoder_blocks

(self, xs: torch.Tensor, xs_masks: torch.Tensor, pos_emb: torch.Tensor, mask_p

wenet/ssl/wav2vec2/wav2vec2_model.py:312

↓ 1 callersMethod_forward_subsampling

( self, xs: torch.Tensor, xs_lens: torch.Tensor )

wenet/ssl/w2vbert/w2vbert_model.py:286

↓ 1 callersMethod_forward_subsampling

( self, xs: torch.Tensor, xs_lens: torch.Tensor )

wenet/ssl/wav2vec2/wav2vec2_model.py:302

↓ 1 callersMethod_get_constant_lr

(self, step)

wenet/utils/scheduler.py:374

↓ 1 callersMethod_get_lr

Simple const lr policy

wenet/utils/scheduler.py:198

↓ 1 callersMethod_get_lr

Simple const lr policy

wenet/utils/scheduler.py:377

↓ 1 callersMethod_get_warmup_lr

(self, step)

wenet/utils/scheduler.py:370

↓ 1 callersFunction_linear_warmup_with_cosine_annealing

(max_lr, warmup_steps, step, decay_steps, min_lr)

wenet/utils/scheduler.py:402

↓ 1 callersFunction_load_json_cmvn

Load the json format cmvn stats file and calculate cmvn Args: json_cmvn_file: cmvn stats file in json format Returns: a num

wenet/utils/cmvn.py:21

↓ 1 callersFunction_load_kaldi_cmvn

Load the kaldi format cmvn stats file and calculate cmvn Args: kaldi_cmvn_file: kaldi text style global cmvn file, which is

wenet/utils/cmvn.py:46

↓ 1 callersFunction_load_paraformer_cmvn

(cmvn_file)

wenet/paraformer/convert_paraformer_to_wenet_config_and_ckpt.py:17

↓ 1 callersMethod_nearest_embedding_idx

(self, xs: torch.Tensor)

wenet/ssl/bestrq/bestrq_model.py:290

↓ 1 callersMethod_noam_annealing

(self, initial_lr, step)

wenet/utils/scheduler.py:616

↓ 1 callersFunction_noam_hold_annealing

(initial_lr, step, warmup_steps, hold_steps, decay_rate, min_lr)

wenet/utils/scheduler.py:438

↓ 1 callersMethod_norm

(self, x)

wenet/transformer/norm.py:19

↓ 1 callersFunction_read_compressed_mat

Read a compressed matrix, see: https://github.com/kaldi-asr/kaldi/blob/master/src/matrix/compressed-matrix.h methods: CompressedMatrix::R

wenet/dataset/kaldi_io.py:487

↓ 1 callersFunction_read_mat_ascii

(fd)

wenet/dataset/kaldi_io.py:472

↓ 1 callersFunction_read_mat_binary

(fd)

wenet/dataset/kaldi_io.py:450

↓ 1 callersMethod_relative_indices

(self, keys: torch.Tensor)

wenet/transformer/attention.py:546

↓ 1 callersFunction_sampler

(pdf: torch.Tensor, num_samples: int, device=torch.device('cpu'))

wenet/ssl/bestrq/mask.py:5

↓ 1 callersMethod_sampler

(self, encoder_out, encoder_out_mask, ys_pad, ys_pad_lens, pre_acoustic_embeds)

wenet/paraformer/paraformer.py:236

↓ 1 callersFunction_square_annealing

(initial_lr, step, max_steps, min_lr)

wenet/utils/scheduler.py:389

↓ 1 callersFunction_squareroot_annealing

(initial_lr, step, max_steps, min_lr)

wenet/utils/scheduler.py:382

↓ 1 callersMethod_stack_features

(self, input: torch.Tensor, input_lens: torch.Tensor)

wenet/ssl/bestrq/bestrq_model.py:257

↓ 1 callersFunction_to_wenet_cmvn

(cmvn_file)

wenet/paraformer/convert_paraformer_to_wenet_config_and_ckpt.py:54

↓ 1 callersFunctionadd_dataset_args

(parser)

wenet/utils/train_utils.py:101

↓ 1 callersFunctionadd_ddp_args

(parser)

wenet/utils/train_utils.py:187

↓ 1 callersFunctionadd_deepspeed_args

(parser)

wenet/utils/train_utils.py:204

↓ 1 callersFunctionadd_fsdp_args

(parser)

wenet/utils/train_utils.py:227

↓ 1 callersFunctionadd_lora_args

Configure parameters for LoRA fine-tuning. Set use_lora and only_optimize_lora to true to enable LoRA functionality. LoRA will be inject

wenet/utils/train_utils.py:126

↓ 1 callersFunctionadd_model_args

(parser)

wenet/utils/train_utils.py:59

↓ 1 callersFunctionadd_trace_args

(parser)

wenet/utils/train_utils.py:89

↓ 1 callersMethodalign

(self, audio_file: str, label: str)

wenet/cli/model.py:156

↓ 1 callersFunctionamp2db

(amp)

wenet/dataset/wav_distortion.py:27

↓ 1 callersFunctionapply_fsdp_checkpointing

(model, ckpt_layer_types: tuple)

wenet/utils/fsdp_utils.py:100

↓ 1 callersFunctionarrange

(s, e, length, keep_length, mask_idc)

wenet/ssl/bestrq/mask.py:112

↓ 1 callersFunctionattention_beam_search

( model, encoder_out: torch.Tensor, encoder_mask: torch.Tensor, beam_size: int = 10, lengt

wenet/transformer/search.py:250

↓ 1 callersFunctionaverage_model_parameters

从step_xxx.pt文件中提取数字，按数字排序后取最后num_pt个文件计算参数平均值，为每个参数保留原始数据类型 Args: pt_files: 模型参数文件路径列表（文件名格式为step_xxx.pt） output_avg_path: 平

common_utils/do_avg_for_exp_dir.py:47

↓ 1 callersFunctionbasic_greedy_search

( model: torch.nn.Module, encoder_out: torch.Tensor, encoder_out_lens: torch.Tensor, n_steps:

wenet/transducer/search/greedy_search.py:6

↓ 1 callersFunctionbatch_backward

(model, scaler, info_dict)

wenet/utils/train_utils.py:726

↓ 1 callersFunctionbatch_inference_llama_questions

处理speech-web-questions_en.jsonl格式文件，输出指定JSON格式结果 Args: jsonl_path: 输入JSONL文件路径 output_json_path: 输出JSON文件路径 prom

infer_code/infer_STTC.py:275