feat(megatron): Qwen3.5 dense + MoE training/inference support via megatron-bridge#1384

Open

Adiactive wants to merge 9 commits into

areal-project:mainfrom

Adiactive:feat/megatron-qwen3_5-pr

Commits on Jun 2, 2026

feat(engine): delegate megatron live weight sync to bridge.export_hf_weights
Adiactive
committed
test(megatron): add Qwen3.5 distributed test scaffolding
Adiactive
committed
feat(engine): add BSHD padded forward path and megatron-bridge patches for Qwen3.5
Adiactive
committed
fix(engine): register Qwen3.5 as vision model and fix non-contiguous broadcast
Adiactive
committed
feat(vllm): add gdn_prefill_backend to avoid FlashInfer GDN hang
Adiactive
committed
test(megatron): add Qwen3.5-MoE expert-parallel + HF save/load tests
Adiactive
committed
docs(example): add Qwen3.5-2B megatron geometry3k GRPO config
Adiactive
committed

Commits on Jun 3, 2026