Skip to content

Latest commit

 

History

History
8 lines (6 loc) · 820 Bytes

README.md

File metadata and controls

8 lines (6 loc) · 820 Bytes



This forked tensorflow repository is used for Parallax. It changes the original tensorflow-v1.6 to keep all the gradients and its target tensors made by automatic differentiation in TensorFlow collection. We also add average_option in SparseConditionalAccumulator to support the diverse average patterns of language models(average_option : Integer indicating how to average out the accumulated values(1: divide by the accumulation count, 2: divide by each element counter, 3: no division). The original tensorflow is here.