feat(models): support TP for qwen3_5_moe - #104

Open

RuixiangMa wants to merge 11 commits into

FlashML-org:mainfrom

RuixiangMa:feat/qwen3-5-tp-support

Commits on Aug 23, 2026

feat(models): support TP for qwen3_5_moe
RuixiangMa
committed
3c8b281

Commits on Aug 26, 2026

feat(models): complete TP for qwen3_5_moe quantized checkpoints (FP8/NVFP4)
CraigStone-Dev
committed
2eae9a8
test: add TP1/TP2 A/B verification scripts for served endpoints
CraigStone-Dev
committed
ab82a3f
fix(models): shard lm_head vocab-parallel in the block-fp8 loader at TP>1
CraigStone-Dev
committed
6197d63
test: tiny block-FP8 checkpoint generator + TP=2 loader regression test
CraigStone-Dev
committed
a77453f
fix(models): shard bf16 lm_head in the mixed loader; never shard NVFP4-native emits
CraigStone-Dev
committed
f1ff8e2

Commits on Aug 31, 2026

fix(moe): partition CPU executor cores per TP rank
CraigStone-Dev
committed
c23ec78

Commits on Sep 2, 2026

Commits on Sep 18, 2026

Merge pull request #1 from CraigStone-Dev/feat/qwen3-5-tp-support
RuixiangMa
authored
e3e6c3d