[#7308] [feat] AutoDeploy: graph-less transformers mode for HF - #7635

Merged

lucaslie merged 25 commits into

NVIDIA:mainfrom

nv-auto-deploy:ll/haoguo/transformers_mode

Sep 18, 2025

Commits on Sep 8, 2025

feat: insert cached attn for transformers mode

h-guo18
authored and
lucaslie
committed
Add transformers.yaml; load weights for factory_model

h-guo18
authored and
lucaslie
committed
minor: clean irrelevant

h-guo18
authored and
lucaslie
committed
address part of review comments;

h-guo18
authored and
lucaslie
committed
address review comments

h-guo18
authored and
lucaslie
committed
add sharding; refine insert_cache by adding profiler;

h-guo18
authored and
lucaslie
committed
polish: use list instead of ptr for shape collection

h-guo18
authored and
lucaslie
committed
feat: flexible cached attn for transformers mode

h-guo18
authored and
lucaslie
committed

Commits on Sep 14, 2025

Merge branch 'main' into ll/haoguo/transformers_mode
lucaslie
authored

Commits on Sep 15, 2025