[7/N] torch.compile, reduce compilation time (#10460)

Signed-off-by: youkaichao <youkaichao@gmail.com>
This commit is contained in:
youkaichao
2024-11-20 11:20:38 -08:00
committed by GitHub
parent 5f1d6af2b6
commit 0cd3d9717e
5 changed files with 27 additions and 16 deletions

View File

@@ -79,7 +79,7 @@ def test_simple_piecewise_compile():
vllm_config = VllmConfig(compilation_config=CompilationConfig(
level=CompilationLevel.PIECEWISE,
use_cudagraph=True,
non_cudagraph_ops=["silly.attention"],
splitting_ops=["silly.attention"],
cudagraph_copy_inputs=True,
))
with set_current_vllm_config(vllm_config):