deepspeedai / deepspeedai/DeepSpeedExamples

How can I find bert weights used for ZeroQuant?

Open
#236 0 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

Dominant language
Python
Stars
6.8k
Forks
1.1k
Avg merge
2d 16h
Merged PRs (30d)
1

Description

Hi I'm trying to reproduce GLUE scores reported on ZeroQuant paper.
But overally almost every tasks' accuracy is lower than reported results in LKD cases.
Escpecially RTE shows sharp degredation in W4/8A8 LKD case.

I assume diffrent results come out because I am using different checkpoint weights
with those used for the research.
The default setting is to get weights from "yoshitomo-matsubara/bert-base-uncased".
How can I get weights file to reproduce at least similar numbers?

Contributor guide

No contributing guide indexed for this repository

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 at the ZeroQuant GLUE evaluation entry point and inspect the default checkpoint setting for yoshitomo-matsubara/bert-base-uncased. Compare the weights used with those from the ZeroQuant research, then rerun the LKD cases, especially RTE W4/8A8; done means achieving results close to the reported GLUE scores.

Written by the indexing model from the issue text.

Assessment

Tech stack
python
Domain
machine-learning
Issue type
Bug
Difficulty
4/5
Estimated time
3-5 days
Activity status
Stale
Clarity
Mostly clear
Newbie friendliness
35/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.