feat: Refactor training framework with modular architecture#3

Open

zzhfz wants to merge 5 commits into

megatron_pr_fixed

Commits on Nov 21, 2025

Refine Megatron script: fix run_id
sunjqa1
committed

Commits on Nov 24, 2025

Fix model selection, pp.type settings, skip warmup, etc.
sunjqa1
committed

Commits on Dec 1, 2025

refactor: replace old megatron runner with modular framework code
sunjqa1
committed

Commits on Dec 3, 2025

refactor: update modular training framework with fixes
sunjqa1
committed

Commits on Dec 5, 2025

feat: add run_id and testcase support with standardized format
sunjqa1
committed