5862708b32
Source: github.com/wudixzy/competition (1527 files, BI-V100 competition reference)
Imported assets:
- 12 prebuilt CoreX .so extensions (corex-3.2.3-ivcore10):
corex_gdn_{beta_decay,causal_conv,gated_norm,packed_decode,qk_map}.so
corex_moe_{direct_routed,exact_reduce,weight_gather}.so
corex_attn_head_rms_norm.so, corex_paged_kv_gather.so
corex_block_major_kv_transfer.so, corex_fused_paged_prefill.so
- 13 CUDA kernel sources (.cu) for above extensions
- 11 build scripts (build_corex_*.sh)
- install_prebuilt_corex.sh (SHA256-verified .so deployment)
- qwen3_5.py (2615 lines) with FULL corex kernel integration
- 9 vllm vendor override files (block manager, sampler, etc)
- 19 patch scripts (model_runner, xformers, block_major, etc)
- Complete serving layer (serving_chat, protocol, api_server, etc)
- bi100_env.py, bi100_profile.py, gdn_prefix.py, block_major_kv_cache.py
- Dockerfile aligned with reference build chain
- computility-run.yaml with BI100_MOE_COREX_DIRECT_ROUTED=1
Call chain verified:
Dockerfile COPY → patch_ops.sh → install_prebuilt_corex.sh → 12 .so to $VLLM_ROOT
qwen3_5.py imports: from vllm import corex_gdn_* / corex_moe_* / corex_attn_*
22 lines
1.8 KiB
Docker
22 lines
1.8 KiB
Docker
FROM harbor.4pd.io/modelhubxc/enginex-iluvatar/bi100-3.2.3-x86-ubuntu20.04-py3.10-poc-llm-infer:v1.2.3
|
|
|
|
ENV PATH=/usr/local/corex/bin:/usr/local/corex-3.2.3/bin:/usr/local/openmpi/bin:${PATH}
|
|
ENV PYTHONPATH=/usr/local/corex/lib64/python3/dist-packages:/usr/local/corex/lib/python3/dist-packages
|
|
ENV LD_LIBRARY_PATH=/usr/local/corex/lib:/usr/local/corex/lib64:/usr/local/corex-3.2.3/lib:/usr/local/corex-3.2.3/lib64:/usr/local/openmpi/lib
|
|
ENV VLLM_ENGINE_ITERATION_TIMEOUT_S=3600 PYTHONUNBUFFERED=1 PYTHONFAULTHANDLER=1 BI100_EXECUTOR_STARTUP_DEBUG=1 ENABLE_CUSTOM_IPC=1
|
|
ENV BI100_PREFIX_MODEL_FINGERPRINT=Qwen3.6-35B-A3B BI100_PREFIX_DTYPE=float16 BI100_PREFIX_TP_SIZE=4
|
|
|
|
RUN mkdir /workspace
|
|
WORKDIR /workspace/
|
|
COPY ./qwen3_6_scripts /workspace/qwen3_6_scripts
|
|
COPY ./vllm/core/evictor_v2.py /workspace/qwen3_6_scripts/vendor_overrides/vllm/core/evictor_v2.py
|
|
COPY ./vllm/core/block/cpu_kv_content_cache.py /workspace/qwen3_6_scripts/vendor_overrides/vllm/core/block/cpu_kv_content_cache.py
|
|
COPY ./vllm/core/block/cpu_gpu_block_allocator.py /workspace/qwen3_6_scripts/vendor_overrides/vllm/core/block/cpu_gpu_block_allocator.py
|
|
COPY ./vllm/core/block/prefix_caching_block.py /workspace/qwen3_6_scripts/vendor_overrides/vllm/core/block/prefix_caching_block.py
|
|
COPY ./vllm/core/block/block_table.py /workspace/qwen3_6_scripts/vendor_overrides/vllm/core/block/block_table.py
|
|
COPY ./vllm/core/block_manager_v2.py /workspace/qwen3_6_scripts/vendor_overrides/vllm/core/block_manager_v2.py
|
|
COPY ./vllm/sampling_params.py /workspace/qwen3_6_scripts/vendor_overrides/vllm/sampling_params.py
|
|
COPY ./vllm/model_executor/sampling_metadata.py /workspace/qwen3_6_scripts/vendor_overrides/vllm/model_executor/sampling_metadata.py
|
|
COPY ./vllm/model_executor/layers/sampler.py /workspace/qwen3_6_scripts/vendor_overrides/vllm/model_executor/layers/sampler.py
|
|
RUN cd ./qwen3_6_scripts && bash ./patch_ops.sh
|