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 2026-08-16 23:39:13 +08:00
cdec569977 feat: 3-tier ixformer flash prefill dispatch + OpenCompass max_tokens clamp + n>1 fanout + index sanitizer
dylanyunlong pushed to main at dylanyunlong/project_6 2026-08-16 23:28:11 +08:00
522e8376b6 data: sub 694 (683分) 日志 + resolve merge
b7149f810a fix: decode MoE路径对齐base — F.linear+bmm替换pre-transpose+bmm
b187f52ced data: so import chain probe
924e48b502 data: so import chain probe
1700c35bd7 test: probe_so_import_chain.sh — 验证.so部署路径+import链+flag值+shape匹配
Compare 9 commits »
dylanyunlong pushed to main at dylanyunlong/project_6 2026-08-16 13:48:16 +08:00
4189f44d27 test: dump ALL symbols from ALL ixformer/cuinfer .so — no grep filter, find what we missed
8eabbac857 test: probe_model_shapes.sh — 真机验证模型config和MoE权重shape
ea2c15f699 fix: decode MoE路径对齐base — F.linear+bmm替换pre-transpose+bmm
ee62ea13ba data: so import chain probe
c290278b35 test: probe_so_import_chain.sh — 验证.so部署路径+import链+flag值+shape匹配
Compare 9 commits »
dylanyunlong pushed to main at dylanyunlong/project_6 2026-08-15 22:56:45 +08:00
b7149f810a fix: decode MoE路径对齐base — F.linear+bmm替换pre-transpose+bmm
dylanyunlong pushed to main at dylanyunlong/project_6 2026-08-15 22:53:59 +08:00
b187f52ced data: so import chain probe
dylanyunlong pushed to main at dylanyunlong/project_6 2026-08-15 22:49:44 +08:00
924e48b502 data: so import chain probe
dylanyunlong pushed to main at dylanyunlong/project_6 2026-08-15 22:48:49 +08:00
1700c35bd7 test: probe_so_import_chain.sh — 验证.so部署路径+import链+flag值+shape匹配
dylanyunlong pushed to main at dylanyunlong/project_6 2026-08-15 22:39:05 +08:00
784dea96c0 data: base MoE forward + corex_moe签名
dylanyunlong pushed to main at dylanyunlong/project_6 2026-08-15 22:37:51 +08:00
adf05b6bfb test: probe_base_moe_forward.sh — cat base qwen3_5.py的完整MoE forward + 所有corex_moe_*.so签名
7a8545f7c2 test: push_probe_results.sh — 真机commit probe结果到modelhub
Compare 2 commits »
dylanyunlong pushed to main at dylanyunlong/project_6 2026-08-15 22:35:09 +08:00
437ad8aaa4 data: probe bridge output
dylanyunlong pushed to main at dylanyunlong/project_6 2026-08-15 22:32:28 +08:00
de1212c271 test: probe_ix_unified_bridge.sh — cat base镜像的ix_unified_bridge + corex_*.so + _custom_ops.py完整接口
dylanyunlong pushed to main at dylanyunlong/project_6 2026-08-15 22:28:33 +08:00
a823bdf9ea test: probe_real_machine.sh — cat ixformer/vllm/cublas真机数据
6415249693 data: port complete MoE + xllm layer call chains from upstream repos
Compare 2 commits »
dylanyunlong pushed to main at dylanyunlong/project_6 2026-08-15 22:28:24 +08:00
7aa5054574 feat: ILU kernel pipeline — ix_full_bridge_v2 build + deploy + 7-step MoE dispatch
52e2ef31a8 feat: xllm_ops NO-FALLBACK kernel loader + 6 missing .so build targets + hot-path patcher
Compare 2 commits »
dylanyunlong pushed to main at dylanyunlong/project_6 2026-08-15 21:15:42 +08:00
e873e5f27b fix: eliminate 8x CUDA sync in MoE decode — tolist() once instead of .item() per expert
dylanyunlong pushed to main at dylanyunlong/project_6 2026-08-15 21:11:38 +08:00
23fe535985 fix: add ex_engine/__init__.py for Python package import
dylanyunlong pushed to main at dylanyunlong/project_6 2026-08-15 21:07:49 +08:00
e18ece8f3a feat: port NaiveBatchedExperts from ds_vllm — view transpose + cublas transB
6f1904aa8c perf: MoE decode — pre-transposed bmm replaces F.linear (6.9ms vs 8.0ms, 14%)
Compare 2 commits »
dylanyunlong pushed to main at dylanyunlong/project_6 2026-08-15 20:47:03 +08:00
9f265894cc test: MoE breakdown — F.linear vs torch.mm vs torch.bmm vs bmm pre-transposed
dylanyunlong pushed to main at dylanyunlong/project_6 2026-08-15 20:45:26 +08:00
e47b66e268 fix: module name in probe_moe_fused_breakdown.sh
dylanyunlong pushed to main at dylanyunlong/project_6 2026-08-15 20:43:32 +08:00
b0af7d54ff test: breakdown moe_decode_fused timing by step — find the real bottleneck
dylanyunlong pushed to main at dylanyunlong/project_6 2026-08-15 20:38:52 +08:00
0795e064b2 fix: corex_batched_gemm TCU OpClassTensorOp + Cu10 + float accum (merge)
3b2a0bc4d3 fix: corex_batched_gemm TCU OpClassTensorOp + Cu10 + float accum
e2fc3f270f fix: corex_batched_gemm use TCU OpClassTensorOp + Cu10 + float accum
d8d241bf9f fix: corex_batched_gemm_kernel — use OpClassTensorOp + arch::Cu10 + FP32 accumulator
Compare 4 commits »