Support uneven tensor-parallel sharding in shard() (GQA/MLA-aware) - #1863

Open

twallgren wants to merge 3 commits into

ml-explore:mainfrom

twallgren:pr3-uneven-tensor-parallel-sharding

Commits on Sep 14, 2026

Convert DeepSeek-V3.2 to PipelineMixin; fix qwen3_moe.py sanitize() gate
twallgren
committed
fed4021
Support uneven tensor-parallel sharding in shard() (GQA/MLA-aware)
twallgren
committed
ab80ec6
Drop stale cache guards already reverted upstream in #1861
twallgren
committed
7ce3051