[Optim][Qwen3TTS][CodePredictor] support torch.compile with reduce-overhead and dynamic False#1913

Merged

hsliuustc0106 merged 11 commits into

vllm-project:mainfrom

JuanPZuluaga:feat/talker-cuda-graph-batched

Mar 20, 2026

Commits on Mar 16, 2026

move to torch.compile reduce-overhead with fixed-shape CUDA graphs
JuanPZuluaga
committed
clean a bit and change setup_compile
JuanPZuluaga
committed
revert docstring
JuanPZuluaga
committed
Merge branch 'main' into feat/talker-cuda-graph-batched
JuanPZuluaga
authored

Commits on Mar 18, 2026