0478628f17
Tested on real machine (cc-b2042074, BI-V100, IX-ML 3.2.3): _moe_C.topk_softmax() → SUCCESS, correct output Two fixes proven on hardware: 1. cuda_compat.h: WARP_SIZE=64 (BI-V100 warp is 64, not 32) 2. topk_softmax_kernels.cu: cub/block/block_reduce.cuh instead of cub/cub.cuh (cub.cuh pulls radix_sort which has WARP_SIZE conflict) Key finding: ixformer SDK on this base image does NOT have topk_softmax. The ixformer::infer namespace from xllm's ixformer.h is for newer SDK. We MUST compile our own _moe_C kernel — which now works. Build flags (clang 16, ivcore10): CUDA: -O3 -cl-fast-relaxed-math (NOT --use_fast_math) C++: -O2 -std=c++17 Dockerfile simplified: 3 steps (was 6) _custom_ops.py: _moe_C as Priority 0, in-place vllm API
25 lines
1.1 KiB
Docker
25 lines
1.1 KiB
Docker
FROM git.modelhub.org.cn:9443/enginex-iluvatar/bi100-3.2.3-x86-ubuntu20.04-py3.10-poc-llm-infer:v1.2.3
|
|
|
|
RUN mkdir -p /workspace
|
|
WORKDIR /workspace/
|
|
|
|
# Copy all sources
|
|
COPY ./qwen3_6_scripts /workspace/qwen3_6_scripts
|
|
COPY ./computility-run.yaml /workspace/computility-run.yaml
|
|
COPY ./ex_engine /workspace/ex_engine
|
|
|
|
# Step 1: Compile _moe_C (CUB-based topk_softmax + moe_align_block_size)
|
|
# Proven on real BI-V100: WARP_SIZE=64, -cl-fast-relaxed-math, cub/block/block_reduce.cuh
|
|
RUN python3 /workspace/ex_engine/precompile_moe_kernels.py 2>&1 | tee /workspace/ex_build.log ; \
|
|
echo "[Dockerfile] _moe_C precompile exit code: $?"
|
|
|
|
# Step 2: Deploy patches (serving + engine fixes)
|
|
RUN chmod +x /workspace/qwen3_6_scripts/patch_ops.sh && \
|
|
bash /workspace/qwen3_6_scripts/patch_ops.sh 2>&1 | tee /workspace/patch_ops.log ; \
|
|
echo "[Dockerfile] patch_ops exit code: $?"
|
|
|
|
# Step 3: Precompile GDN kernel (needs vllm in path, so after patch_ops)
|
|
RUN python3 /workspace/qwen3_6_scripts/precompile_gdn.py \
|
|
/workspace/qwen3_6_scripts/flash_qla_sm70 2>&1 | tee -a /workspace/ex_build.log ; \
|
|
echo "[Dockerfile] gdn precompile exit code: $?"
|