deepspeedai / deepspeedai/DeepSpeed

Converting DeepSpeedTransformerLayer to Hugging Face Transformers' BERTLayer

Open
#1,028 0 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

Dominant language
Python
Stars
43.1k
Forks
5k
Avg merge
4d 15h
Merged PRs (30d)
112

Description

Hi, thank you for publicizing this amazing work!
When pre_layer_norm=False, I think that the DeepSpeedTransformerLayer is exactly equivalent to the vanilla hidden layer of the transformer architecture. If so, is it possible to convert DeepSpeedTransformerLayer to BERTLayer in Hugging Face Transformers?
It seems that DeepSpeedTransformerLayer has the following parameters: 'attn_qkvw', 'attn_qkvb', 'attn_ow', 'attn_ob', 'attn_nw', 'attn_nb', 'inter_w', 'inter_b', 'output_w', 'output_b', 'norm_w', 'norm_b'. When assigning them to the corresponding weights in BERTLayer, can we obtain the same model?

Contributor guide

Open the contributing guide

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

Research direction

Start by comparing DeepSpeedTransformerLayer with Hugging Face Transformers' BERTLayer, using the listed parameters as the mapping checklist. Determine whether assigning those weights produces an equivalent model, and document the conversion steps or any incompatibilities found.

Written by the indexing model from the issue text.

Assessment

Tech stack
huggingface, python, pytorch
Domain
machine-learning
Issue type
Feature
Difficulty
5/5
Estimated time
Over a week
Activity status
Stale
Clarity
Needs clarification
Newbie friendliness
25/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.