bojone / t5_in_bert4keras

整理一下在keras中使用T5模型的要点
171 stars 28 forks source link

載入T5m模型ckpt時,出現『not an sstable (bad magic number)』錯誤!! #2

Closed fox88-tw closed 4 years ago

fox88-tw commented 4 years ago

RuntimeError Traceback (most recent call last) /usr/local/lib/python3.6/dist-packages/tensorflow/python/training/py_checkpoint_reader.py in NewCheckpointReader(filepattern)

RuntimeError: not an sstable (bad magic number)

During handling of the above exception, another exception occurred:

OpError Traceback (most recent call last) 9 frames /usr/local/lib/python3.6/dist-packages/tensorflow/python/training/py_checkpoint_reader.py in error_translator(e)

OpError: not an sstable (bad magic number)`

載入路徑 config_path = '/content/drive/My Drive/T5m_Pre-model/mt5_base_config.json' checkpoint_path = '/content/drive/My Drive/T5m_Pre-model/model.ckpt-1000000' spm_path = '/content/drive/My Drive/T5m_Pre-model/sentencepiece_cn.model' keep_tokens_path = '/content/drive/My Drive/T5m_Pre-model/sentencepiece_cn_keep_tokens.json'

網路上除錯方式,皆指向修改ckpt檔名,將model.ckpt後贅詞刪除即可。並不適用,因早已參照劍大code上所寫的,將".data-00000-00016"贅詞移除;且即使將"-1000000"移除,還是會發生同樣問題。

版本: google colab tensorflow==1.14.0 bert4keras==0.9.1 sentencepiece==0.1.94

fox88-tw commented 4 years ago

我找到 問題點了!! 是我下載 ckpt 檔時,並沒有下載完全所導致.....

dzapologize commented 1 year ago

大佬,请问去哪里下载ckpt可以下载完整的model.ckpt-1000000呀?我这边服务器和网盘都是下载的model.ckpt-1000000.data-00000-of-00016到model.ckpt-1000000.data-00015-of-00016