Delete check_2.py

check_2.py

@@ -1,100 +0,0 @@
-import argparse# run.py
-import os
-import sys
-import json
-import torch
-from tqdm import tqdm
-import numpy as np
-import re
-import google.genai as genai
-
-def count_lengths(input_lengths):
-    input_lengths = (input_lengths - 1) // 2 + 1
-    output_lengths = (input_lengths - 2) // 2 + 1
-    return input_lengths, output_lengths
-
-def load_audio_qwenomni(audio_path):
-    audio = whisper.load_audio(audio_path, sr=16000).tolist()
-    if len(audio) % 160 != 0:
-        audio += [0] * (160 - len(audio) % 160)
-    audio = np.array(audio, dtype=np.float32)
-    mel = whisper.log_mel_spectrogram(audio, n_mels=128)
-    len_feature = mel.shape[1]
-    input_len, output_len = count_lengths(len_feature)
-
-    audio_info = {
-        "audio_feats": mel,
-        "lens_audio_feats": [len_feature],
-        "input_lens": [input_len],
-        "output_lens": [output_len]
-    }
-    return audio_info
-
-whisper = __import__("whisper")
-
-import warnings
-# 忽略特定类型的警告，这里针对 UserWarning
-warnings.filterwarnings("ignore", category=UserWarning)
-
-
-from transformers import AutoConfig, Qwen2_5OmniForConditionalGeneration
-
-
-def load_audio(audio_path, audio_encoder, device):
-     with torch.no_grad():
-      audio_info = load_audio_qwenomni(audio_path)
-      audio_feats = torch.tensor(audio_info["audio_feats"]).to(device)
-      input_feat_lens = torch.tensor(audio_info["lens_audio_feats"]).to(device)
-      input_lens = torch.tensor(audio_info["input_lens"]).to(device)
-      audio_feat = audio_encoder(audio_feats, input_feat_lens, input_lens).last_hidden_state
-
-      if audio_feat.size(1) != 1280:
-            raise ValueError("audio_feat size is not 1280")
-
-     return audio_feat
-
-def process_jsonl(jsonl_path, device, audio_encoder):
-      new_feat_savepath = os.path.join(os.path.dirname(jsonl_path), 'audio_feats')
-      if not os.path.exists(new_feat_savepath):
-           os.makedirs(new_feat_savepath)
-
-      bad_case_file = jsonl_path.replace('.jsonl', '_bad_case.jsonl')
-
-      print(f"Processing {jsonl_path} on GPU {device}")
-      with open(jsonl_path, "r") as f:
-          lines = f.readlines()
-      for line in tqdm(lines):
-            try:
-                  data_item = json.loads(line)
-                  old_feats_path = data_item["audio_feat_path"]
-                  old_wav_path = os.path.join('/mnt/home/xiezhifei/projects/data_machine3', old_feats_path).replace('/audio_feats/', '/audio/').replace('.pt', '.wav')
-                  file_name = os.path.basename(old_feats_path)
-                  audio_feat = load_audio(old_wav_path, audio_encoder, device)
-                  torch.save(audio_feat, os.path.join(new_feat_savepath, file_name))
-            except Exception as e:
-                  line['error'] = e
-                  with open(bad_case_file, 'a') as f:
-                        f.write(json.dumps(line, ensure_ascii=False) + '\n')
-                  
-BASE_FOLDER = "/mnt/home/xiezhifei/projects/data_machine3/tts_data_0720_machine3_tobe_compressed"
-
-def main():
-      parser = argparse.ArgumentParser()
-      parser.add_argument('--split_id', type=str, required=True)
-      parser.add_argument('--gpu_id', type=int, required=True)
-      args = parser.parse_args()
-
-      device = f"cuda:{args.gpu_id}"
-
-      qwen_omni_config = AutoConfig.from_pretrained("/mnt/home/xiezhifei/projects/zh/checkpoint/stage1_paskale2e/qwen2.5-omni-3B")
-      audio_encoder =  Qwen2_5OmniForConditionalGeneration._from_config(qwen_omni_config).thinker.audio_tower
-      audio_encoder.load_state_dict(torch.load("/mnt/home/xiezhifei/projects/zh/download/datasets--zh-liu799--passsss/snapshots/cac004a84c5bcb68fe9b4136c5b6f191159998c0/audio_tower_adapterfinetuned_0615.pth", map_location=device))
-      audio_encoder.requires_grad_(False).eval
- 
-      audio_encoder.to(device)
-      from glob import glob
-      jsonl_path = glob(f"{BASE_FOLDER}/tts_{args.split_id}/*.jsonl")[0]
-      process_jsonl(jsonl_path, device, audio_encoder)
-
-if __name__ == "__main__":
-    main()