Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Adds script for AWQ-quantizing model #101

Open
wants to merge 3 commits into
base: main
Choose a base branch
from

Conversation

ru5h16h
Copy link

@ru5h16h ru5h16h commented Mar 4, 2024

Pull Request Checklist

Reference Issue

ref: castorini/ura-projects#4

Checklist Items

Before submitting your pull request, please review these items:

  • [Yes] Have you followed the contributing guidelines?
  • [Yes] Have you verified that there are no existing Pull Requests for the same update/change?
  • [Yes] Have you updated any relevant documentation or added new tests where needed?

PR Type

What kind of change does this PR introduce?

  • Bugfix
  • Feature
  • Code style update (formatting, local variables)
  • Refactoring (no functional changes, no api changes)
  • Documentation content changes
  • Other...
    • Description: Adds model quantization using AWQ

@ru5h16h ru5h16h marked this pull request as draft March 4, 2024 00:23
save_awq.py Outdated
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This should not be in the main repo folder, probably src/rank_llm/scripts

Copy link
Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

ack

Copy link
Member

@sahel-sh sahel-sh left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Thank you for the PR!

save_awq.py Outdated
parser.add_argument(
"--dataset",
type=str,
default="msp_open_ai_ada2_random_s5000_gpt4_da0_mr20_sampled_mix.jsonl",
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

is it possible to have rank zephy's training data or a subset of it as the default value of the calibration dataset?

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This would probably requires changes to the load dataset logic too

Copy link
Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This is the file that @ronakice shared. Wasn't this one used for training?

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

No, this is not the data that we used for finetuning rankzephyr, but I leave it to Ronak to decide if we want to you the training dataset or the one that we shared with you.

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@ronakice PTAL!

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

/u3/rpradeep/RankVicuna/data/msp_open_ai_ada2_random_s5000_gpt4_da0_mr20_sampled_mix.jsonl

This is the file I used to train RankZephyr @sahel-sh?

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

either way, something is off in AWQ quantizing, i will advice against merging until this is properly sorted

save_awq.py Outdated Show resolved Hide resolved
Copy link
Member

@sahel-sh sahel-sh left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

sorry for being slow on this, LGTM!

@ru5h16h ru5h16h marked this pull request as ready for review April 5, 2024 20:04
@ru5h16h
Copy link
Author

ru5h16h commented Apr 5, 2024

Here are the details outlining the insights gathered and other experimental information: https://docs.google.com/document/d/1BHpN9lDVGjtjIAFMxjUxNuu1K4KJOgIaOZXkWF1_K8c/edit

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

3 participants