Code

GPT-Neo ONNX export#12911

Code

Merged

michaelbenayoun merged 11 commits intohuggingface:masterhuggingface/transformers:masterfrom michaelbenayoun:onnx_gpt_neomichaelbenayoun/transformers:onnx_gpt_neoCopy head branch name to clipboardAug 5, 2021

+380-103

Commits

Commits on Jul 26, 2021

• onnx export for GPT-Neo

  michaelbenayoun

  committed
  8bdf05b

  Copy full SHA for 8bdf05b

Commits on Jul 27, 2021

• task support for ONNX export

  michaelbenayoun

  committed
  92c9e85

  Copy full SHA for 92c9e85

Commits on Jul 28, 2021

• custom_unfold implementation without Python dynamic control flow / objects (draft version with lots of printing and comments, comitted to have them available if need be)

  michaelbenayoun

  committed
  21d4641

  Copy full SHA for 21d4641

• custom_unfold implementation without Python dynamic control flow / objects

  michaelbenayoun

  committed
  d51a30e

  Copy full SHA for d51a30e

• fixed TYPE_HINT issue in __init__.py

  michaelbenayoun

  committed
  9068eb2

  Copy full SHA for 9068eb2

Commits on Jul 29, 2021

• fixing vectorized implementation of unfold, and implemented GPTNeoAttentionMixin._get_block_length_and_num_blocks in a graph friendly fashion

  michaelbenayoun

  committed
  425f59a

  Copy full SHA for 425f59a

Commits on Jul 30, 2021

• support for use_past

  michaelbenayoun

  committed
  2d3ce2f

  Copy full SHA for 2d3ce2f

Commits on Aug 2, 2021

• task / feature refactoring

  michaelbenayoun

  committed
  fd421dc

  Copy full SHA for fd421dc

Commits on Aug 4, 2021

• task / feature refactoring (part 2), and support for sequence-classification

  michaelbenayoun

  committed
  237bc0b

  Copy full SHA for 237bc0b

• test fix

  michaelbenayoun

  committed
  1347095

  Copy full SHA for 1347095

• documented PatchingSpec

  michaelbenayoun

  committed
  da26af8

  Copy full SHA for da26af8