dylanyunlong/project_6
1
0
Fork 0
Code Issues Pull Requests Actions Projects Releases Wiki Activity

Commit Graph

  • 9ca33cf4d5 upstream: add GEMM kernel references from 4 repos for BI-V100 porting Claude 2026-08-14 15:11:57 +00:00
  • 29ecc2e602 feat: moe_expert_gemm.cpp — C++ loop over experts via ixformer_linear (replaces Python for-loop) Claude 2026-08-14 12:09:50 +00:00
  • 0ace44e293 test: cat ixinfer.h + functional batched GEMM test on BI-V100 Claude 2026-08-14 12:07:00 +00:00
  • bfc4de2cf3 test: probe cuinfer + cublasLt + cublas batched GEMM APIs for MoE expert dispatch Claude 2026-08-14 12:03:52 +00:00
  • d6958070cb test: probe CUTLASS + __CUDA_ARCH__ on BI-V100 for grouped GEMM feasibility Claude 2026-08-14 11:59:51 +00:00
  • 50a249e0a3 Revert "feat: batched MoE expert GEMM — replaces Python for-loop" claude 2026-08-14 11:47:37 +00:00
  • 06d7713db6 feat: batched MoE expert GEMM — replaces Python for-loop claude 2026-08-14 11:43:46 +00:00
  • 93353a1414 test: probe ixformer .so symbols for MoE grouped GEMM claude 2026-08-14 11:40:13 +00:00
  • 865c18f852 feat: integrate xllm_moe into qwen3_5.py MoE hot path claude 2026-08-14 11:37:21 +00:00
  • e147c283e3 prebuilt: 5 xllm CUDA kernel .so for BI-V100 (ivcore10) root 2026-08-14 11:34:23 +00:00
  • 31d3ee99bb fix: MoE kernel include paths — device_utils.cuh + arch_condition.h claude 2026-08-14 11:31:54 +00:00
  • df6a0f5d47 fix: remove cuda/functional from MoE topk kernels (not available on corex) claude 2026-08-14 11:29:19 +00:00
  • a50adefdfc feat: xllm MoE CUDA kernels — fused_topk + compute_index + combine claude 2026-08-14 11:23:49 +00:00
  • 49cd7def89 fix: slot_ids dtype long→int32 in cache test claude 2026-08-14 11:20:07 +00:00
  • 1f51feee05 fix: cos_sin_cache dtype float→half in rope test claude 2026-08-14 11:18:02 +00:00
  • 7fc0c1defa test: AST call chain verification for all 4 xllm kernel .so claude 2026-08-14 11:14:58 +00:00
  • 3d816cd18d fix: add ceil_div + DEVICE_INLINE to device_utils.cuh claude 2026-08-14 11:11:15 +00:00
  • 302aa9608a fix: block_copy.cu — DEVICE_INLINE, CHECK_EQ→TORCH_CHECK, cstdint claude 2026-08-14 11:06:07 +00:00
  • 900ae0b1ef fix: block_copy.cu remove utils.h (glog), CHECK→TORCH_CHECK claude 2026-08-14 11:02:57 +00:00
  • a206fc1d43 fix: activation.cu torch/extension.h + LOG(FATAL)→TORCH_CHECK, reshape_paged_cache.cu torch header claude 2026-08-14 10:52:54 +00:00
  • 093bfb380f feat: pybind11 bindings for xllm CUDA kernels claude 2026-08-14 10:50:13 +00:00
  • 415fff85f1 fix: add DISPATCH_FLOATING_TYPES macro to device_utils.cuh claude 2026-08-14 10:45:37 +00:00
  • 0359103b9b fix: remove glog/cuda_ops_api.h dependency from all xllm CUDA kernels claude 2026-08-14 10:30:10 +00:00
  • 51cb90b9ab fix: adapt xllm norm.cu for corex CUB (CUDA 10.2) claude 2026-08-14 10:19:51 +00:00
  • 089b9ff4e2 test: fix CUB compat tests + probe corex built-in CUB claude 2026-08-14 10:11:10 +00:00
  • 793743f5c0 test: CUB compatibility on BI-V100 — warp shuffle + BlockReduce claude 2026-08-14 08:21:48 +00:00
  • ec140f3605 docs: complete porting assessment — 5 repos, 2660 build targets, ivcore10 compatibility claude 2026-08-14 08:17:25 +00:00
  • 3a2cfc87c9 test: xllm CUDA kernel verification suite for BI-V100 claude 2026-08-14 08:01:23 +00:00
  • 8d75652949 feat: import CUDA kernels from xllm/CCCL/FLA upstream repos claude 2026-08-14 07:48:52 +00:00
  • 051b02d3cd feat: ix_moe_bridge + ix_attn_bridge — dlopen bridges for full ixformer::infer API claude 2026-08-14 07:32:31 +00:00
  • 5e9b7c292a build: all 16 .so prebuilt, no JIT compile in docker Claude 2026-08-14 07:14:46 +00:00
  • 28102196cd Merge branch 'main' of https://github.com/dylanyunlon/project_6 root 2026-08-14 07:13:36 +00:00
  • d32822c5d2 prebuilt: corex_moe_index_combine.so + ix_full_bridge.so (BI-V100 compiled) root 2026-08-14 07:10:52 +00:00
  • e7247bd57b build: integrate ix_full_bridge + corex_moe_index_combine into docker patch_ops.sh Claude 2026-08-14 07:09:01 +00:00
  • d9ffc5159d fix(bridge): rewrite ix_full_bridge.cpp for real ixformer_torch_ext namespace Claude 2026-08-14 07:00:30 +00:00
  • 9e3157b444 fix(P0): extra=allow + topk_softmax fallback + deploy_local.sh + SO chain verify Claude 2026-08-14 06:56:00 +00:00
  • 101db8774c diag: probe_bi100.py — run on real machine, paste output back Claude 2026-08-14 06:52:52 +00:00
  • 07fad2cce7 fix: max-num-seqs=1 + pure Q-tiling — match a3c45d3b exactly Claude 2026-08-14 05:40:05 +00:00
  • 1f69311375 fix: flash_attn import path ixformer.contrib → ixformer.functions Claude 2026-08-14 03:56:23 +00:00
  • 4c1adc11db fix: restore libcccl_allocator.so + minimal .dockerignore + clean pycache Claude 2026-08-14 03:51:43 +00:00
  • dfaaae988e diag: test_dlopen_chain.py — verify all .so and ixformer dispatch on BI-V100 Claude 2026-08-14 03:47:39 +00:00
  • 336f3349ca fix(submit): restore flash_attn prefill + all 38eca5c2 improvements Claude 2026-08-14 03:47:09 +00:00
  • 67a5639c3c fix(submit): restore a3c45d3b-proven config + safe improvements Claude 2026-08-14 03:35:24 +00:00
  • bce79e44be perf(prefill): restore flash_attn_varlen_func with profiling safety guard Claude 2026-08-14 02:51:47 +00:00
  • 74ce61712b fix(build): update prebuilt CoreX artifact count 13→14 — SHA256SUMS has 14 entries after corex_gdn_chunk_recurrent.so was added Claude 2026-08-14 02:39:42 +00:00
  • 38eca5c26a revert: restore a3c45d3b yaml + Q-tiling + remove all OOM hacks Claude 2026-08-14 02:28:48 +00:00
  • eb57eb7d1c clean: remove 444 .pyc files + libcccl_allocator.so from git tracking Claude 2026-08-14 02:20:26 +00:00
  • 41b51382fd prebuilt: corex_gdn_chunk_recurrent.so + remove __pycache__ from tracking root 2026-08-14 02:12:32 +00:00
  • 9a52f05783 prebuilt: add corex_gdn_chunk_recurrent.so with fixed pybind kwargs root 2026-08-14 02:10:54 +00:00
  • bb0de83d45 fix(pybind): c10::optional<Tensor>() default — c10::nullopt not registered in pybind Claude 2026-08-14 02:05:30 +00:00
  • 768d89c31a fix(pybind): add py::arg + defaults to corex_gdn_chunk_recurrent Claude 2026-08-14 02:02:51 +00:00
  • a4d16d36b8 fix(build): std::nullopt → c10::nullopt for CoreX c10::optional Claude 2026-08-14 01:50:47 +00:00
  • 56fe58ada3 fix(build): c10::optional for CoreX CUDA 10.2 — std::optional incompatible Claude 2026-08-14 01:49:25 +00:00
  • 1d9b620416 fix(crash): disable corex_gdn_chunk_recurrent — pybind signature mismatch Claude 2026-08-14 01:36:14 +00:00
  • 716034bdd0 fix(OOM): lower blocks cap 5000→3000 — flash_attn needs ~4GB temp buffer project6-dev 2026-08-14 01:33:06 +00:00
  • c8a982c4e8 feat: ix_full_bridge.so — dlopen bridge for ixformer::infer C++ API Claude 2026-08-14 01:25:16 +00:00
  • 872be0effa fix(build): strip \r\n from all .py files — CRLF breaks patch_ops.sh text matching Claude 2026-08-14 01:06:49 +00:00
  • aa4b4992d1 fix(build): hardcode blocks cap 5000 in .py — remove yaml env var project6-dev 2026-08-14 01:00:09 +00:00
  • 456380eed0 fix(OOM): cap GPU blocks at 5000 via BI100_MAX_GPU_BLOCKS env var project6-dev 2026-08-14 00:12:39 +00:00
  • c6aa1b9c62 fix(P0): protocol.py extra=allow — recover 180 rejected replay requests Claude 2026-08-14 00:10:54 +00:00
  • 20aac5b212 fix(OOM): return zeros during profiling — skip both flash_attn AND Q-tiling project6-dev 2026-08-13 16:36:17 +00:00
  • 048302bd4a Revert "fix: remove --enable-chunked-prefill — conflicts with small max_num_batched_tokens" project6-dev 2026-08-13 16:35:49 +00:00
  • 15ad56a454 fix: remove --enable-chunked-prefill — conflicts with small max_num_batched_tokens project6-dev 2026-08-13 16:34:44 +00:00
  • e31bd69779 fix(OOM): max-num-batched-tokens 4096→256 — profiling Q-tiling also OOMs project6-dev 2026-08-13 16:13:41 +00:00
  • 2717bafc30 fix(OOM): skip flash_attn_varlen during profiling project6-dev 2026-08-13 15:30:25 +00:00
  • aebc660a10 revert: restore to 8c8c0286 (last confirmed build success) project6-dev 2026-08-13 15:08:40 +00:00
  • bed1fc4d54 fix(OOM): skip flash_attn_varlen during profiling — use Q-tiling fallback project6-dev 2026-08-13 14:41:40 +00:00
  • 4518a39d12 fix: restore LD_PRELOAD libcccl_allocator.so — verified compiles in docker build project6-dev 2026-08-13 14:13:19 +00:00
  • 8c8c0286c9 Revert "fix(build): revert patch_ops.sh to a3c45d3b — remove cccl_preload + corex extension compile steps" project6-dev 2026-08-13 13:56:13 +00:00
  • 14d1725cdd fix(build): revert patch_ops.sh to a3c45d3b — remove cccl_preload + corex extension compile steps project6-dev 2026-08-13 13:53:21 +00:00
  • 09d92dce5d fix(build): remove 1106 extra CCCL headers — keep only 288 needed by allocator project6-dev 2026-08-13 13:40:17 +00:00
  • 5ec60dc574 fix(build): remove --num-gpu-blocks-override — may fail platform yaml validation project6-dev 2026-08-13 13:34:27 +00:00
  • 71644e1530 fix: num-gpu-blocks-override 4000→5500 — 4000 blocks can't fit a 70K prompt project6-dev 2026-08-13 13:25:23 +00:00
  • cf7824313f fix(OOM): add --num-gpu-blocks-override=4000 to skip profiling project6-dev 2026-08-13 13:24:36 +00:00
  • 8d2f30f065 fix(critical): remove LD_PRELOAD libcccl_allocator.so — crashes service if .so missing Claude 2026-08-13 13:15:13 +00:00
  • 451bdc8204 fix: lower gpu-memory-utilization to 0.85 — prevent profiling OOM Claude 2026-08-13 12:38:42 +00:00
  • 6092edebde fix: import torch.utils.cpp_extension explicitly in build script project6-dev 2026-08-13 11:37:10 +00:00
  • f6cf9d662e fix(CCCL): split compilation to isolate CCCL headers from torch/corex project6-dev 2026-08-13 11:35:43 +00:00
  • 05706f0d60 fix(build): use block-level CUB only — device-level API conflicts with corex CUDA 10.2 project6-dev 2026-08-13 11:25:37 +00:00
  • daa8067080 fix: remove expandable_segments:True from PYTORCH_CUDA_ALLOC_CONF Claude 2026-08-13 11:22:32 +00:00
  • 4c365b8c03 feat(CCCL): device-level CUB algorithms for MoE dispatch project6-dev 2026-08-13 11:18:52 +00:00
  • 7ba97f7977 fix: relax .dockerignore — remove *.txt *.md *.json wildcards that may break build Claude 2026-08-13 10:46:31 +00:00
  • 45161610f0 fix: thread_local reentrant guard — prevent cudaMalloc infinite recursion Claude 2026-08-13 10:37:32 +00:00
  • 3ce5bff10f fix: use cccl_preload::cub namespace — CUB_WRAPPED_NAMESPACE requires it Claude 2026-08-13 10:36:27 +00:00
  • c1e23615b5 fix: remove CUB_WRAPPED_NAMESPACE and _CCCL_COMPILER_GCC from build flags Claude 2026-08-13 10:35:11 +00:00
  • 32325f9624 build: wire CCCL preload into competition pipeline dylanyunlon 2026-08-13 10:31:17 +00:00
  • 3af2a32eb5 fix: verify_submission.sh — strip semicolons from path extraction grep Claude 2026-08-13 10:21:23 +00:00
  • 9ef5af3bda fix: wire CCCL preload into build+launch chain + pre-submission verification Claude 2026-08-13 10:00:26 +00:00
  • a6b5891bfc feat: CCCL CachingDeviceAllocator preload — 完整依赖链 288 files dylanyunlon 2026-08-13 09:24:42 +00:00
  • 8dc6462a2b feat: CCCL CachingDeviceAllocator LD_PRELOAD — bypass CoreX expandable_segments ASSERT dylanyunlon 2026-08-13 09:21:57 +00:00
  • 887e0981ad fix: remove expandable_segments — CoreX CUDACachingAllocator不支持 dylanyunlon 2026-08-13 09:08:47 +00:00
  • 9cfc6c72c0 feat(CCCL): LD_PRELOAD CachingDeviceAllocator — intercept cudaMalloc/cudaFree Claude 2026-08-13 09:21:45 +00:00
  • ca42633148 build: trigger rebuild after merge reconciliation root 2026-08-13 08:46:38 +00:00
  • 7e4e04b7c6 build: trigger rebuild after merge reconciliation root 2026-08-13 08:46:38 +00:00
  • 089e810984 feat: CCCL CachingDeviceAllocator preload — 完整依赖链 288 files dylanyunlon 2026-08-13 09:24:42 +00:00
  • e7c703ef94 feat: CCCL CachingDeviceAllocator LD_PRELOAD — bypass CoreX expandable_segments ASSERT dylanyunlon 2026-08-13 09:21:57 +00:00
  • c1e7065076 fix: remove expandable_segments — CoreX CUDACachingAllocator不支持 dylanyunlon 2026-08-13 09:08:47 +00:00
  • 1ea2100cb8 feat(CCCL): LD_PRELOAD CachingDeviceAllocator — intercept cudaMalloc/cudaFree Claude 2026-08-13 09:21:45 +00:00
  • abbc13c4d5 build: trigger rebuild after merge reconciliation root 2026-08-13 08:46:38 +00:00
  • a8304bf906 build: trigger rebuild after merge reconciliation root 2026-08-13 08:46:38 +00:00