Files
enginex-mlu370-vllm/vllm-v0.6.2/tests/compile/piecewise/piecewise_compilation_config.json
2026-02-04 17:22:39 +08:00

5 lines
110 B
JSON

{
"use_cudagraph": true,
"non_cudagraph_ops": ["silly.attention"],
"cudagraph_copy_inputs": true
}