ancient_rain
0 Followers · 0 Following
  • Joined on 2026-07-14
Repositories
1
 Projects Packages Public Activity Starred Repositories
ancient_rain created pull request ccf-ai-infra/TileOPs-Metax#55 2026-08-06 09:37:50 +08:00
[moe_reduce_fused] feat: 实现并优化 C500 专家输出加权归约算子
ancient_rain closed pull request ccf-ai-infra/TileOPs-Metax#48 2026-08-06 09:36:01 +08:00
#[moe_reduce_fused] feat: 实现并优化 C500 专家输出加权归约算子
ancient_rain pushed to feat/moe_reduce_fused at ancient_rain/TileOPs-Metax 2026-08-06 09:07:04 +08:00
ff401eed69 test(moe): fold official matrix into workloads/moe.py per review
ancient_rain pushed to feat/moe_reduce_fused at ancient_rain/TileOPs-Metax 2026-08-06 08:42:56 +08:00
ancient_rain pushed to feat/moe_reduce_fused at ancient_rain/TileOPs-Metax 2026-08-06 08:40:25 +08:00
ff401eed69 test(moe): fold official matrix into workloads/moe.py per review
ancient_rain pushed to feat/moe_reduce_fused at ancient_rain/TileOPs-Metax 2026-08-06 08:01:09 +08:00
f75e22bc89 fix(moe): compare fp8 outputs in fp32 in official matrix assert_equal
1756245247 fix(moe): derive _infer_output_shapes from input shapes for manifest --strict parity
Compare 2 commits »
ancient_rain pushed to feat/moe_reduce_fused at ancient_rain/TileOPs-Metax 2026-08-06 07:24:18 +08:00
ancient_rain pushed to feat/moe_reduce_fused_gy at ancient_rain/TileOPs-Metax 2026-08-06 07:21:31 +08:00
483cc32a35 docs(moe): archive PR B results materials on gy branch
ancient_rain pushed to feat/moe_reduce_fused at ancient_rain/TileOPs-Metax 2026-08-06 07:20:26 +08:00
f2b617e7e7 chore(moe): remove PR B results docs folder from merge branch
ancient_rain created pull request ccf-ai-infra/TileOPs-Metax#48 2026-08-06 00:59:12 +08:00
#[moe_reduce_fused] feat: 实现并优化 C500 专家输出加权归约算子
ancient_rain closed pull request ccf-ai-infra/TileOPs-Metax#46 2026-08-06 00:54:30 +08:00
[moe_reduce_fused] feat: 新增 TileLang Kernel、4 个 Op 变体、测试与 Benchmark（PR B）
ancient_rain pushed to feat/moe_reduce_fused at ancient_rain/TileOPs-Metax 2026-08-06 00:53:00 +08:00
63f22ad720 docs(moe): align PR B baseline narrative with TileLang baseline
4e40dae770 docs(moe): switch PR B baseline to TileLang kernel comparison
Compare 2 commits »
ancient_rain created pull request ccf-ai-infra/TileOPs-Metax#46 2026-08-05 23:59:51 +08:00
[moe_reduce_fused] feat: 新增 TileLang Kernel、4 个 Op 变体、测试与 Benchmark（PR B）
ancient_rain pushed to feat/moe_reduce_fused at ancient_rain/TileOPs-Metax 2026-08-05 23:57:33 +08:00
ea29be63be docs(moe): record tested commit SHA in PR B report
ancient_rain pushed to feat/moe_reduce_fused at ancient_rain/TileOPs-Metax 2026-08-05 23:54:37 +08:00
be764cc463 docs(moe): add PR B report, benchmark logs, and C500 correctness evidence
ancient_rain pushed to feat/moe_reduce_fused at ancient_rain/TileOPs-Metax 2026-08-05 19:48:15 +08:00
7770740ac4 test(moe): cover qwen3-235b/qwen3-30b shapes in reduce-fused variant tests
0e811542b1 feat(moe): add qwen3 workloads to MoeReduceFused scaled variants
Compare 2 commits »
ancient_rain pushed to feat/moe_reduce_fused at ancient_rain/TileOPs-Metax 2026-08-05 18:35:04 +08:00
2d92c86c15 fix(moe): correct roofline FLOPs for MoeReduceFused scaling variants
ancient_rain pushed to feat/moe_reduce_fused at ancient_rain/TileOPs-Metax 2026-08-04 18:20:56 +08:00
a409c2d18b fix(moe): drop torch.compile in official matrix data gen
db9e68d40d Merge feat/moe_reduce_fused_zhb into feat/moe_reduce_fused_gy
71e36dd735 fix(moe): tolerance assert, env-gated matrix, and official-matrix test guide
dbc355bd13 add manifest workloads
d2562e3537 test(moe): add official TileKernels matrix from codex branch
Compare 8 commits »
ancient_rain pushed to feat/moe_reduce_fused_gy at ancient_rain/TileOPs-Metax 2026-08-04 17:45:19 +08:00
a409c2d18b fix(moe): drop torch.compile in official matrix data gen
ancient_rain pushed to feat/moe_reduce_fused_gy at ancient_rain/TileOPs-Metax 2026-08-04 17:36:57 +08:00
db9e68d40d Merge feat/moe_reduce_fused_zhb into feat/moe_reduce_fused_gy
dbc355bd13 add manifest workloads
Compare 2 commits »