Go to file

dylan ddcfbad431 feat: pybind wrapper for CUTLASS batched GEMM → MoE decode path

Based on verified result (issue #68):
  CUTLASS Cu10 TensorOp batched: 2.462ms (8 experts, 1 launch)
  vs 8× torch.matmul: 4.6ms (8 launches)
  vs Python F.linear loop: 10.36ms

New files:
  ex_engine/xllm_kernels/cuda/bindings/corex_batched_gemm_bind.cpp
    pybind11 wrapper: batched_gemm_fp16() + moe_decode_fused()
  ex_engine/xllm_kernels/cuda/corex_batched_gemm_kernel.cu
    CUTLASS GemmBatched<half> kernel (from cat_files/batched_gemm.cu)
  qwen3_6_scripts/build_corex_batched_gemm.sh
    Build script for BI-V100 (ivcore10)

Modified:
  qwen3_6_scripts/qwen3_5.py
    import corex_batched_gemm + _USE_COREX_BATCHED_GEMM flag
    Tier 1.5 in MoE decode: after corex_direct_routed, before corex_gather

Build on device: bash qwen3_6_scripts/build_corex_batched_gemm.sh
Output: prebuilt/corex-3.2.3-ivcore10/corex_batched_gemm.so

2026-08-15 11:54:26 +00:00

cat_files

Revert "data: cat SGEMM files from 3 repos into cat_files/"

2026-08-15 11:48:21 +00:00

cccl_upstream

test: xllm CUDA kernel verification suite for BI-V100

2026-08-14 08:01:23 +00:00

docs

docs: complete porting assessment — 5 repos, 2660 build targets, ivcore10 compatibility

2026-08-14 08:17:25 +00:00

ex_engine

feat: pybind wrapper for CUTLASS batched GEMM → MoE decode path

2026-08-15 11:54:26 +00:00

ixformer_sdk

feat(CRITICAL): 从 GitHub 扫描搬运 ixformer SDK + xllm 完整 GDN/MoE 代码

2026-08-11 02:32:06 +00:00

muh

test: add scan tuning verification — union SMEM model from agent_scan.cuh

2026-08-07 03:16:44 +00:00

optimizations

fix: critical config + tuning corrections from CCCL source analysis

2026-08-03 06:45:54 +00:00

qwen3_6_scripts

feat: pybind wrapper for CUTLASS batched GEMM → MoE decode path

2026-08-15 11:54:26 +00:00

upstream_ref

data: complete SGEMM upstream from 3 repos (siboehm+wangzyon+edtallison) + xllm fused_qknorm_rope + xattention kernels

2026-08-15 07:00:09 +00:00

vllm

feat: import CUDA kernels from xllm/CCCL/FLA upstream repos

2026-08-14 07:48:52 +00:00

vllm_adapter

feat: add qwen3_5.py vllm adapter for Qwen3.6-35B-A3B

2026-08-01 13:54:33 +00:00

vllm_overrides

feat: import CUDA kernels from xllm/CCCL/FLA upstream repos

2026-08-14 07:48:52 +00:00

__init__.py

[DEPLOY] Complete submission: baseline + all optimizations

2026-07-30 16:06:20 +00:00

.dockerignore

fix: restore libcccl_allocator.so + minimal .dockerignore + clean pycache

2026-08-14 03:51:43 +00:00

.gitattributes

fix(build): strip \r\n from all .py files — CRLF breaks patch_ops.sh text matching

2026-08-14 01:06:49 +00:00

.gitignore

prebuilt: corex_gdn_chunk_recurrent.so + remove __pycache__ from tracking

2026-08-14 02:12:32 +00:00

attention.py

[ENGINE] attention.py: CCCL dispatch_reduce.cuh single-tile decision for V1/V2

2026-08-07 01:54:40 +00:00

baseline.muh

fix(critical): disable thinking for tool_call requests — fixes d03_tool_call FAIL

2026-08-07 07:45:28 +00:00

BI_V100_BENCHMARK_RUNBOOK.md

[docs] BI-V100 benchmark runbook: Phase 0 硬件探测 + Phase 1 quick sweep + Phase 3 端到端验证

2026-08-04 01:14:30 +00:00

cat_ixformer_vllm.py

perf: native ixformer decode (v1 ≤32K, v2 >32K) + flash_attn_varlen prefill

2026-08-13 07:04:21 +00:00

CCCL_ASSET_MAP.md

[CCCL] Add missing CCCL components: c2h, nvbench_helper, cmake, cudax, AGENTS.md

2026-08-06 02:14:18 +00:00

CCCL_INTEGRATION_STATUS.md

docs: CCCL integration status — 462 assets inventory, 8/8 decode hot path coverage, no-clone-more verdict

2026-08-03 03:50:24 +00:00

CCCL_MUH_GAP_ANALYSIS.md

analysis: 26-algorithm CCCL↔muh full gap scan — 19.7% coverage, 299 benchmark points needed, 3/26 READY

2026-08-07 02:43:04 +00:00

CCCL_MUH_PARITY_AUDIT.md

audit: CCCL vs muh parity check — scale_mem_bound 11/11 PASS, SM100 benchmark values extracted

2026-08-07 07:19:00 +00:00

CCCL_PATTERN_MAP.md

docs: CCCL → vllm kernel pattern mapping for competition

2026-08-07 03:19:39 +00:00

cccl_sm100_benchmark_values.json

audit: CCCL vs muh parity check — scale_mem_bound 11/11 PASS, SM100 benchmark values extracted

2026-08-07 07:19:00 +00:00

CCCL_TUNING_GAP_REPORT.md

[analysis] CCCL↔muh 26-algorithm tuning gap report — 294 bench pts needed, 19% line coverage, reduce/scan/topk P0

2026-08-07 01:45:37 +00:00

chat_dataset_v0.json

[DEPLOY] Complete submission: baseline + all optimizations

2026-07-30 16:06:20 +00:00

CODEPATH_MAP.md

doc: add functional test FAIL root cause analysis to CODEPATH_MAP

2026-08-09 00:03:45 +00:00

COMP168_DIAGNOSIS.md

diag: comp 168 完整分析 + verify_deployment.sh

2026-08-12 03:08:35 +00:00

COMPETITIVE_ANALYSIS_AND_FIX_PLAN.md

fix(critical): disable thinking for tool_call requests — fixes d03_tool_call + d05/t5 FAIL

2026-08-07 08:45:22 +00:00

computility-run.fix.yaml

fix(build): 回退qwen3_6_scripts+ex_engine到26e6cb40(能得分版本)

2026-08-12 01:33:24 +00:00

computility-run.ref.yaml

feat(CRITICAL): import wudixzy/competition complete corex stack — 12 prebuilt .so + 13 CUDA kernels + 2615-line qwen3_5.py

2026-08-11 03:55:38 +00:00

computility-run.yaml

fix: max-num-seqs=1 + pure Q-tiling — match a3c45d3b exactly

2026-08-14 05:40:05 +00:00

computility-run.yaml.bak

fix(build): 回退qwen3_6_scripts+ex_engine到26e6cb40(能得分版本)

2026-08-12 01:33:24 +00:00

debug_gdn_nan.py

perf: native ixformer decode (v1 ≤32K, v2 >32K) + flash_attn_varlen prefill

2026-08-13 07:04:21 +00:00

debug_topk.py

debug: topk kernel mismatch diagnostic

2026-08-10 07:59:06 +00:00

debug_warpsize.py

debug: check BI-V100 warp size

2026-08-10 08:02:19 +00:00

deltanet_chunk_optimize.py

[CCCL-PORT] Two architecture-level optimizations from CCCL system design

2026-08-05 08:20:01 +00:00

DEVELOPMENT_STATUS.md

fix(CRITICAL): engine death on image request + stop overwriting base corex modules

2026-08-10 09:12:05 +00:00

diagnose_build.sh

fix(CRITICAL): Docker build 全步容错 + patch_ops.sh函数定义顺序修复

2026-08-11 16:39:10 +00:00

DLOPEN_DEV_PLAN.md

perf: native ixformer decode (v1 ≤32K, v2 >32K) + flash_attn_varlen prefill

2026-08-13 07:04:21 +00:00

DLOPEN_DISPATCH_CHAIN.md

feat(EX): corex_gdn + corex_moe — dlopen dispatch chain from comp 168 log analysis

2026-08-10 03:37:15 +00:00

Dockerfile

fix(build): match project_7 proven Dockerfile — remove ENV lines + .dockerignore

2026-08-13 02:48:28 +00:00

Dockerfile.broken_head

fix(CRITICAL): merge 26e6cb40 build pipeline + HEAD features — fix docker build

2026-08-11 07:58:14 +00:00

Dockerfile.broken_head2

fix(build): 回退qwen3_6_scripts+ex_engine到26e6cb40(能得分版本)

2026-08-12 01:33:24 +00:00

Dockerfile.fix

fix(build): 回退qwen3_6_scripts+ex_engine到26e6cb40(能得分版本)

2026-08-12 01:33:24 +00:00

Dockerfile.ref

feat(CRITICAL): import wudixzy/competition complete corex stack — 12 prebuilt .so + 13 CUDA kernels + 2615-line qwen3_5.py

2026-08-11 03:55:38 +00:00

dockerrizhi.txt

comp 168 docker

2026-08-07 08:43:51 +00:00

engine_cccl_patterns.py

[ENGINE] Pattern 7: CCCL C API JIT → Triton autotune mapping

2026-08-06 02:32:34 +00:00

ENGINE_CODEPATH_TIMELINE.md

arch(CoreX): CCCL env_dispatch — try native fused kernels, fallback PyTorch

2026-08-08 11:15:04 +00:00

ENGINEX_INJECTION_MAP.md

docs: EngineX vllm injection map — Python not C++, Triton not CUDA, 32KB SMEM claim

2026-08-03 03:59:49 +00:00

enginex-vllm-bi100-qwen36-main.zip

update baseline

2026-07-30 15:19:37 +00:00

GROUND_TRUTH_STATUS_v2.md

[DOC] GROUND_TRUTH_STATUS v2 — based on complete code reading

2026-08-06 02:21:03 +00:00

GROUND_TRUTH_STATUS.md

[STATUS] Ground truth: gen_patch C++ injection dead, real path is Triton+Python tuning

2026-08-05 07:57:53 +00:00

HARDWARE_PROBE_20260808.md

doc: BI-V100 hardware probe raw data (SSH Aug 8 2026)

2026-08-08 18:18:48 +00:00

launch_service

sync: align with enginex-vllm-bi100-qwen36 baseline (1902c81f)

2026-07-31 09:43:58 +00:00

muh_cc_dispatch.py

[ENGINE] muh_cc_dispatch + analysis: max_num_seqs=1 from computility-run.yaml

2026-08-06 01:02:21 +00:00

muh_dispatch.py

[muh_dispatch] 修正 head_dim=128→256 + 删除死代码 + 强制 V1

2026-08-05 07:12:26 +00:00

muh_kernel_map.py

[docs+code] lookback delay is a no-op on BI-V100 + V2 compound reduce pattern

2026-08-05 03:35:07 +00:00

MUH_PROJECT_CHECKPOINT.md

[MUH] Fix 7 structural discrepancies vs CCCL — read source, not grep

2026-07-30 14:37:38 +00:00

MUH_TUNING_GAP_ANALYSIS.md

docs: comprehensive CCCL vs muh gap analysis with competition priority

2026-08-07 03:18:25 +00:00

paged_attention_v2_pytorch.py

[ENGINE] paged_attention_v2: CCCL single-tile fast path + GridEvenShare constants

2026-08-06 02:21:29 +00:00

paged_attention_v2_triton.py

[ARCH] CCCL-derived paged attention kernel architecture + Triton rewrite

2026-07-31 04:13:07 +00:00

paged_attn.py

fix(critical): raise decode threshold to prevent service crash

2026-08-07 04:39:19 +00:00

PIPELINE_GROUND_TRUTH.md

docs: pipeline ground truth — CCCL parity audit, SM100 benchmark data, injection status

2026-08-07 07:20:34 +00:00

PIPELINE_REALITY_CHECK.md

feat: add gen_config.py (Python-layer config generator) + pipeline reality check

2026-08-07 03:11:56 +00:00

PIPELINE_STATUS.md

docs: add PIPELINE_STATUS.md — ground truth for muh injection mapping and toolchain status

2026-08-07 01:45:46 +00:00

PRD.md

doc(prd): complete base engine migration checklist

2026-08-08 10:48:43 +00:00

prefix_prefill.py

[prefill] asymmetric BLOCK_M/BLOCK_N from CCCL AgentReduce insight

2026-08-05 03:07:20 +00:00

probe_all_so.py

fix(bridge): ix_moe_bridge链接加--unresolved-symbols + probe脚本

2026-08-11 09:48:14 +00:00

probe_all_symbols.sh

debug: probe all ixformer symbol locations

2026-08-10 06:33:14 +00:00

probe_base_moe.py

debug: probe base image vllm FusedMoE actual dispatch chain

2026-08-10 06:41:52 +00:00

probe_bi100.py

diag: probe_bi100.py — run on real machine, paste output back

2026-08-14 06:52:52 +00:00

probe_ixformer_symbols.py

fix: fail-fast on ix_bridge failure + probe script for real machine

2026-08-10 03:04:50 +00:00

probe_kv_layout2.py

perf: native ixformer decode (v1 ≤32K, v2 >32K) + flash_attn_varlen prefill

2026-08-13 07:04:21 +00:00

probe_kv_layout.py

perf: native ixformer decode (v1 ≤32K, v2 >32K) + flash_attn_varlen prefill

2026-08-13 07:04:21 +00:00

probe_moe_detail.py

debug: deep probe MoE kernel dispatch in base image

2026-08-10 07:13:01 +00:00

probe_paged_attn.py

perf: native ixformer decode (v1 ≤32K, v2 >32K) + flash_attn_varlen prefill

2026-08-13 07:04:21 +00:00

probe_symbol.sh

debug: probe script to find silu_and_mul symbol location

2026-08-10 06:32:04 +00:00

PROJECT_SUMMARY.md

docs: update PROJECT_SUMMARY — comp 168 analysis + three critical fixes

2026-08-10 06:56:59 +00:00

README.md

Initial commit

2026-07-30 17:03:23 +08:00

SO_BUILD_MANIFEST.md

data: complete SGEMM upstream from 3 repos (siboehm+wangzyon+edtallison) + xllm fused_qknorm_rope + xattention kernels

2026-08-15 07:00:09 +00:00

SUB509_DEEP_DIAGNOSIS.md

fix(deltanet): systematic NaN elimination via CCCL overflow_cast pattern

2026-08-07 09:26:11 +00:00

SUB509_DIAGNOSIS.md

docs: complete sub508 diagnosis — 19 cascade failures from n=2, full test matrix

2026-08-07 07:54:09 +00:00

subrizhi.txt

update sub 655

2026-08-13 02:29:40 +00:00

SYSTEM_DESIGN.md

doc: system design — architecture, file map, data flow, build pipeline

2026-08-10 02:07:43 +00:00

test_dlopen_chain.py

fix: flash_attn import path ixformer.contrib → ixformer.functions

2026-08-14 03:56:23 +00:00

test_ex_engine_cuda.py

test: add test_ex_engine_cuda.py — test all 22 prebuilt .so on BI-V100

2026-08-15 06:28:34 +00:00

test_triton.py

perf: native ixformer decode (v1 ≤32K, v2 >32K) + flash_attn_varlen prefill

2026-08-13 07:04:21 +00:00

TUNING_SURFACE_TRUTH.md

fix: revert invalid patches, add honest tuning surface assessment

2026-08-03 10:34:28 +00:00

verify_build.sh

test: verify_build.sh for real hardware validation

2026-08-11 07:22:18 +00:00

verify_deployment.sh

diag: comp 168 完整分析 + verify_deployment.sh

2026-08-12 03:08:35 +00:00

verify_docker_build.sh

fix: Dockerfile Step 7 VLLM_ROOT污染 — vllm INFO日志混入stdout

2026-08-11 09:28:21 +00:00

verify_flash_attn.py

perf: native ixformer decode (v1 ≤32K, v2 >32K) + flash_attn_varlen prefill

2026-08-13 07:04:21 +00:00

verify_forward.py

fix: total_mem → total_memory

2026-08-11 13:18:53 +00:00

verify_gdn_cpp.py

perf: native ixformer decode (v1 ≤32K, v2 >32K) + flash_attn_varlen prefill

2026-08-13 07:04:21 +00:00

verify_ixformer_attn.py

perf: native ixformer decode (v1 ≤32K, v2 >32K) + flash_attn_varlen prefill

2026-08-13 07:04:21 +00:00

verify_moe_e2e.py

perf: native ixformer decode (v1 ≤32K, v2 >32K) + flash_attn_varlen prefill

2026-08-13 07:04:21 +00:00

verify_moe_index_combine.py

perf: native ixformer decode (v1 ≤32K, v2 >32K) + flash_attn_varlen prefill

2026-08-13 07:04:21 +00:00

verify_on_device.py

test: verify_on_device.py — 真机逐函数验证, 不允许fallback

2026-08-10 04:06:21 +00:00

verify_paged_attn.py

perf: native ixformer decode (v1 ≤32K, v2 >32K) + flash_attn_varlen prefill

2026-08-13 07:04:21 +00:00

verify_single_card.py

fix: total_mem → total_memory (corex torch API)

2026-08-10 07:55:39 +00:00

verify_single_gpu.py

refactor(bridge): rewrite ix_full_bridge.cpp for actual base image symbols

2026-08-10 06:34:54 +00:00

verify_so_loading.py

diag: verify_so_loading.py — check all 13 prebuilt .so + base image dlopen chain

2026-08-13 01:45:36 +00:00

verify_submission.sh

fix: verify_submission.sh — strip semicolons from path extraction grep

2026-08-13 10:21:23 +00:00

verify_topk_softmax.py

perf: native ixformer decode (v1 ≤32K, v2 >32K) + flash_attn_varlen prefill

2026-08-13 07:04:21 +00:00

README.md

project_6

Languages

C++ 41.8%

Cuda 31.6%

Python 22.2%

C 2.1%

CMake 1.1%

Other 1.1%