From 27878c86899c51ad82f56d7b881012900b9d9c47 Mon Sep 17 00:00:00 2001 From: DP Migration Date: Wed, 2 Sep 2026 01:00:37 +0000 Subject: [PATCH] under test, not sure no errors --- qwen3_6_scripts/ex_engine/build_xllm_kernels.sh | 12 ++++++++++++ 1 file changed, 12 insertions(+) diff --git a/qwen3_6_scripts/ex_engine/build_xllm_kernels.sh b/qwen3_6_scripts/ex_engine/build_xllm_kernels.sh index 23c57c8e..675644c4 100755 --- a/qwen3_6_scripts/ex_engine/build_xllm_kernels.sh +++ b/qwen3_6_scripts/ex_engine/build_xllm_kernels.sh @@ -148,6 +148,18 @@ echo "============================================================" build_so "xllm_moe" \ "ex_engine/xllm_kernels/cuda/moe/moe_fused_topk.cu ex_engine/xllm_kernels/cuda/moe/moe_compute_index.cu ex_engine/xllm_kernels/cuda/moe/moe_combine.cu ex_engine/xllm_kernels/cuda/bindings/xllm_moe_bind.cpp" +# 7. py_attention_metadata — DP-aware attention metadata (pybind11, pure C++) +# Source: core/runtime/py_attention_metadata.cpp +# Ported from xLLM upstream PR #2258 (commit 78aa2a85) +# Exports: AttentionMetadataView with dp_token_counts / dp_is_decode fields +# Note: Pure C++ pybind11 — no CUDA, no .cu files +echo "" +echo "============================================================" +echo " 7. py_attention_metadata.so" +echo "============================================================" +build_so "py_attention_metadata" \ + "core/runtime/py_attention_metadata.cpp" + echo "" echo "============================================================" echo " Build complete. Output:"