feat(EX): ix_full_bridge — all 14 ixformer::infer functions bridged
Upstream source: xllm/core/kernels/ilu/ixformer.h (Apache 2.0)
Wrapper patterns: xllm/core/kernels/ilu/{attention,norm,rope,activation,fused_moe,group_gemm}.cpp
Complete bridge (ix_full_bridge.cpp, 331 lines):
MoE: topk_softmax, gen_idx, expand, group_gemm, silu_mul, combine, fused_forward
Attention: paged_attention (decode), flash_attn_prefill (prefill)
Norm: rms_norm, fused_add_rms_norm
RoPE: rotary_embedding
Cache: reshape_and_cache
Linear: ixformer_linear
ix_bridge.py: tries ix_full_bridge first, falls back to ix_moe_bridge
patch_ops.sh: deploys both .cpp files to all JIT search paths
Copied ixformer.h + utils.h headers for reference
This commit is contained in:
@@ -197,11 +197,13 @@ if [ -d "$EX_ENGINE_SRC/python" ]; then
|
||||
EX_DST="$VLLM/model_executor/models/ex_engine"
|
||||
mkdir -p "$EX_DST/python" "$EX_DST/csrc"
|
||||
cp "$EX_ENGINE_SRC/python/"*.py "$EX_DST/python/" 2>/dev/null || true
|
||||
# ix_moe_bridge.cpp for JIT compile — deploy to ALL search paths
|
||||
cp "$EX_ENGINE_SRC/csrc/ix_moe_bridge.cpp" "$EX_DST/csrc/" 2>/dev/null || true
|
||||
cp "$EX_ENGINE_SRC/csrc/ix_moe_bridge.cpp" "$EX_DST/python/" 2>/dev/null || true
|
||||
cp "$EX_ENGINE_SRC/csrc/ix_moe_bridge.cpp" "/workspace/ex_engine/csrc/" 2>/dev/null || true
|
||||
cp "$EX_ENGINE_SRC/csrc/ix_moe_bridge.cpp" "/workspace/qwen3_6_scripts/" 2>/dev/null || true
|
||||
# ix_full_bridge.cpp + ix_moe_bridge.cpp for JIT compile — deploy to ALL search paths
|
||||
for _BRIDGE in ix_full_bridge.cpp ix_moe_bridge.cpp; do
|
||||
cp "$EX_ENGINE_SRC/csrc/$_BRIDGE" "$EX_DST/csrc/" 2>/dev/null || true
|
||||
cp "$EX_ENGINE_SRC/csrc/$_BRIDGE" "$EX_DST/python/" 2>/dev/null || true
|
||||
cp "$EX_ENGINE_SRC/csrc/$_BRIDGE" "/workspace/ex_engine/csrc/" 2>/dev/null || true
|
||||
cp "$EX_ENGINE_SRC/csrc/$_BRIDGE" "/workspace/qwen3_6_scripts/" 2>/dev/null || true
|
||||
done
|
||||
touch "$EX_DST/__init__.py"
|
||||
touch "$EX_DST/python/__init__.py"
|
||||
# Copy built .so files
|
||||
|
||||
Reference in New Issue
Block a user