eval dataloader and sampler changes

feat: add example gemma3 config

fix: temp remove liger unknown model patch

remove debug logs and simplify

see if it passes e2e

chore(docs): add cookbook/blog link to docs

fix: add validation for liger for nonsupported models

fixing sample packing

tracking the main

working multi-group SP

e2e test debug - transformers version

only validate hf user token on rank 0

e2e test for llama flex attnetion w/ sample packing

typo

requirements: transformers on main

add pydantic validation for attention fields

rebased flex attn integration

remove reference to deprecated import

fix: handle system role properly depending on template

fix: format

fix: format

fix: use tagged gemma3 transformers

fix: rearrange sort order

feat: add gemma2 e2e test

test import-wihtin-import relative path

Built site for gh-pages

Built site for gh-pages

use max of 32 dataset processes if not explicit (#2403)