dylanyunlong

0 Followers · 0 Following

• Joined on 2026-08-06

Repositories

10

Projects Public Activity Starred Repositories

dylanyunlong pushed to main at dylanyunlong/project_6 2026-08-13 17:35:40 +08:00

19946d4c72 merge: reconcile modelhub with github (keep CUB_NS_QUALIFIER fix)

188d3654ef build: trigger rebuild after merge reconciliation

0ac118911d fix: CUB_NS_QUALIFIER for wrapped namespace + drop _CCCL_COMPILER_GCC

8d6f9eaeb0 feat: CCCL CachingDeviceAllocator preload — 完整依赖链 288 files

967d572073 feat: CCCL CachingDeviceAllocator LD_PRELOAD — bypass CoreX expandable_segments ASSERT

Compare 7 commits »

dylanyunlong pushed to main at dylanyunlong/project_6 2026-08-13 17:30:48 +08:00

e1cdbfc167 feat: CCCL CachingDeviceAllocator preload — 完整依赖链 288 files

8eba8df925 feat: CCCL CachingDeviceAllocator LD_PRELOAD — bypass CoreX expandable_segments ASSERT

1a8d086c4a fix: remove expandable_segments — CoreX CUDACachingAllocator不支持

d036646726 feat(CCCL): LD_PRELOAD CachingDeviceAllocator — intercept cudaMalloc/cudaFree

Compare 4 commits »

dylanyunlong pushed to main at dylanyunlong/project_6 2026-08-13 16:46:54 +08:00

05713aa7cb build: trigger rebuild after merge reconciliation

dylanyunlong pushed to main at dylanyunlong/project_6 2026-08-13 15:10:45 +08:00

a1ae6e366f merge: reconcile squashed commit with modelhub history

d2b4df54ff perf: native ixformer decode — v1 ≤32K, v2 >32K (no Python fallback)

f28223c9da perf: native ixformer decode (v1 ≤32K, v2 >32K) + flash_attn_varlen prefill

Compare 3 commits »

dylanyunlong pushed to main at dylanyunlong/project_6 2026-08-13 14:40:42 +08:00

e78fa560c8 feat: wire corex_gdn_chunk_recurrent C++ kernel into GDN prefill path

c720cbc3a3 docs: dlopen SO开发计划 — gap analysis from comp168 log + real tree

Compare 2 commits »

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

17fdf7e2d6 diag: probe KV cache layout with 5D key + 4D value

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

cb03fc9993 diag: cat ixformer vllm.py source

dylanyunlong pushed to main at dylanyunlong/project_6 2026-08-13 13:19:26 +08:00

b0ed88e114 diag: probe ixformer KV cache 5D layout + read vllm.py source

dylanyunlong pushed to main at dylanyunlong/project_6 2026-08-13 13:18:15 +08:00

a8f0332e1c diag: verify_paged_attn.py — test ixformer paged attention v1/v2 with head_dim=256

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

ce568f94ed diag: probe ixformer paged attention signature

ad6863ed84 perf: replace Python Q-tiling fallback with ixformer.flash_attn_varlen_func

Compare 2 commits »

dylanyunlong pushed to main at dylanyunlong/project_6 2026-08-13 13:11:33 +08:00

9f02200ede diag: verify_flash_attn.py — test flash_attn_func correctness + perf for head_dim=256

dylanyunlong pushed to main at dylanyunlong/project_6 2026-08-13 12:28:31 +08:00

9c97a24edf diag: verify_ixformer_attn.py — test ixformer native attention with head_dim=256

dylanyunlong pushed to main at dylanyunlong/project_6 2026-08-13 12:19:02 +08:00

1aa2262a2c diag: test_triton.py — check if Triton works on BI-V100

dylanyunlong pushed to main at dylanyunlong/project_6 2026-08-13 12:15:50 +08:00

a3f223ae45 fix: correct module name in debug_gdn_nan.py

dylanyunlong pushed to main at dylanyunlong/project_6 2026-08-13 12:12:58 +08:00

a617b743a3 diag: debug_gdn_nan.py — isolate NaN source in C++ chunk GDN

dylanyunlong pushed to main at dylanyunlong/project_6 2026-08-13 12:09:15 +08:00

0861de65d0 feat: C++ GDN chunk+recurrent from xllm upstream + verification script

dylanyunlong pushed to main at dylanyunlong/project_6 2026-08-13 11:57:03 +08:00

6f7d25f26d fix: add .dockerignore (exclude __pycache__) + e2e MoE verification script

dylanyunlong pushed to main at dylanyunlong/project_6 2026-08-13 11:54:38 +08:00

796b09952c feat: integrate moe_compute_index kernel into MoE prefill path

dylanyunlong pushed to main at dylanyunlong/project_6 2026-08-13 11:50:35 +08:00

71d39a1c7e feat: moe_compute_index + moe_combine_result CUDA kernels from xllm upstream

3045f29814 fix: enable corex topk_softmax kernel — verified correct on real BI-V100

Compare 2 commits »

dylanyunlong pushed to main at dylanyunlong/project_6 2026-08-13 11:26:57 +08:00

fc7a089334 diag: verify_topk_softmax.py — test kernel vs PyTorch on real BI-V100

0b0c47fddd fix(critical): fold max_completion_tokens + max_num_seqs=2 + max_model_len=80000 + xllm_latest layer import

Compare 2 commits »