Support original z-lab Qwen3 DFlash checkpoints; require checkpoint target_layer_ids - #4991

Open

harshal-96 wants to merge 3 commits into

InternLM:mainfrom

harshal-96:feat/dflash-qwen3-dense-rebased

Commits on Sep 19, 2026

feat: support original z-lab Qwen3 DFlash checkpoints and mixed dtypes
harshal-96
committed
3c8ccdf
remove build_target_layer_ids fallback; require checkpoint target_layer_ids
harshal-96
committed
5794d5a

Commits on Sep 20, 2026

address review: mixed-dtype tests, top-level mask_token_id test, clearer block_size error
harshal-96
committed
57883c2