dylanyunlong

0 Followers · 0 Following

• Joined on 2026-08-06

Repositories

6

Projects Public Activity Starred Repositories

dylanyunlong pushed to main at dylanyunlong/project_6 2026-08-15 20:05:37 +08:00

3481f2903f fix(build): cutlass.h lives under tensorflow/include on this image

f41900c06b fix(build): auto-find cutlass/cutlass.h under COREX_ROOT

Compare 2 commits »

dylanyunlong pushed to main at dylanyunlong/project_6 2026-08-15 20:01:32 +08:00

bfa18cd5b4 fix(build): use CoreX clang++ instead of nvcc — match working build scripts

dylanyunlong pushed to main at dylanyunlong/project_6 2026-08-15 19:59:22 +08:00

04cc9b88af fix(build): add CUDA include path to g++ step in build_corex_batched_gemm.sh

dylanyunlong pushed to main at dylanyunlong/project_6 2026-08-15 19:57:25 +08:00

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

a875fa5d4c Revert "data: cat SGEMM files from 3 repos into cat_files/"

36676f2d1b data: complete SGEMM upstream from 3 repos (siboehm+wangzyon+edtallison) + xllm fused_qknorm_rope + xattention kernels

7cfa87b5ac data: cat SGEMM files from 3 repos into cat_files/

284804ac53 data: cat 3 SGEMM repos — siboehm, wangzyon, edtallison (full clone, no --depth)

Compare 7 commits »

dylanyunlong pushed to main at dylanyunlong/project_6 2026-08-15 13:58:41 +08:00

109d29fa60 Merge remote-tracking branch 'modelhub/main'

045ea5df79 feat: Cu10 TensorOp batched HGEMM via Iluvatar CUTLASS framework

Compare 2 commits »

dylanyunlong pushed to main at dylanyunlong/project_6 2026-08-15 13:41:31 +08:00

30f98c0674 data: Cu10 CUTLASS part 2 — tensorop example, arch.h, cutlass.h

dylanyunlong pushed to main at dylanyunlong/project_6 2026-08-15 13:41:22 +08:00

f006ab1a01 test: cat tensorop GEMM example + arch.h + cutlass.h from corex-samples

dylanyunlong pushed to main at dylanyunlong/project_6 2026-08-15 13:32:13 +08:00

b922d694dc data: Cu10 CUTLASS headers from corex-samples

dylanyunlong pushed to main at dylanyunlong/project_6 2026-08-15 13:31:52 +08:00

1d36754efc fix: cat_cutlass_cu10.sh writes to cat_files/ directory instead of stdout

4abb4df215 test: cat Cu10 CUTLASS files — mma_cu10.h, iluvatar_mma.hpp, batched_gemm.cu, default_mma_core_cu10.h

Compare 2 commits »

dylanyunlong pushed to main at dylanyunlong/project_6 2026-08-15 13:28:05 +08:00

6b9086c3a9 test: probe Cu10 CUTLASS fork — find mma_cu10.h, tensor op files, batched_gemm example

dylanyunlong pushed to main at dylanyunlong/project_6 2026-08-15 13:22:52 +08:00

a465dd1d75 fix: use F.silu in test script for old corex torch

a12d070d82 fix: replace torch::silu with x*sigmoid(x) for old corex torch

Compare 2 commits »

dylanyunlong pushed to main at dylanyunlong/project_6 2026-08-15 13:20:21 +08:00

c840c9159f feat: moe_tcu_dispatch.cpp — C++ MoE expert loop via torch::mm (TCU kernel)

dylanyunlong pushed to main at dylanyunlong/project_6 2026-08-15 13:17:32 +08:00

9514092980 test: probe torch.matmul backend + ixformer.matmul/linear + Python loop overhead

dylanyunlong pushed to main at dylanyunlong/project_6 2026-08-15 13:14:35 +08:00

395b3e4042 test: clean rebuild + debug output for kernel 10 correctness

dylanyunlong pushed to main at dylanyunlong/project_6 2026-08-15 13:12:59 +08:00

a8ca42b59c perf: hgemm_warptiling Config B — beats cublas on MoE-sized GEMM (0.7x)

dylanyunlong pushed to main at dylanyunlong/project_6 2026-08-15 13:08:36 +08:00

21417319bc test: sweep 6 kernel 10 configs + cublas baseline — find best params for warp64

dylanyunlong pushed to main at dylanyunlong/project_6 2026-08-15 01:24:15 +08:00

27bb8d28df test: probe kernel 10 perf with CUDA events — isolate bottleneck

dylanyunlong pushed to main at dylanyunlong/project_6 2026-08-15 01:06:28 +08:00

2b12fe687e feat: hgemm_warptiling.cu — siboehm kernel 10 ported to WARPSIZE=64 FP16

dylanyunlong pushed to main at dylanyunlong/project_6 2026-08-15 01:03:23 +08:00

11b8a98eea test: probe warp_size=64 behavior + kernel 10 warp tiling with WARPSIZE=64 on BI-V100

dylanyunlong pushed to main at dylanyunlong/project_6 2026-08-15 00:54:18 +08:00

1af7e7cf48 fix: use c10::cuda::getCurrentCUDAStream().stream() for corex torch