Skip to content

Re-producing the results for the full parameter models #7

@kxk302

Description

@kxk302

Hi,
I am trying to reproduce the results in Table 2, specifically, the results for "Full-Rank" row in the table for 60m parameter model.

Image

I am running scripts/llm_pretrain/sltrain60m.sh, but I have modified --peft_model from "sltrain" to "full". After 11,000 steps, I get this:

"Eval loss and perplexity at step 11001: 3.4039955139160156, 30.084061520394627"

Whereas the perplexity reported in Table 2 is 34.06

Could you please advise on how to re-produce the Table 2 results? Greatly appreciate your help!

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions