ray_chan
/
TileOPs-Metax
forked from ccf-ai-infra/TileOPs-Metax
1
0
Fork 1
Code Issues Pull Requests Actions 3 Packages Projects Releases Wiki Activity

Commit Graph

  • 1b2620103c docs(skills): register-resident GPU memory-hierarchy decision rule feat/quant_swiglu_fwd_channel_cast_transpose ray24777 2026-08-05 11:57:53 +0000
  • 3b980a4f09 refactor(quant_swiglu): drop stage-profiling variants and their scripts ray24777 2026-08-05 11:56:40 +0000
  • 22ba38ceae test(quant_swiglu): enforce bit-exact fp8 contract with torch ray24777 2026-08-05 11:35:44 +0000
  • d886342ad6 optimize(quant_swiglu): swizzled act_shared layout for nt=128 transposed (-9 to -13% on nt128-transposed) ray24777 2026-08-05 08:38:18 +0000
  • 8eb099ec36 optimize(quant_swiglu): register-resident non-transposed tile, no act_shared (-12.8% sum) ray24777 2026-08-05 06:00:04 +0000
  • 9166434743 optimize(quant_swiglu): patch-mapped read + register absmax (5.9% sum) ray24777 2026-08-05 01:39:17 +0000
  • 7dbf963d36 optimize(quant_swiglu): config-tunable tile/thread/stages with full validation ray24777 2026-08-04 10:06:08 +0000
  • 7c77c9d1d1 optimize(quant_swiglu): hand-written fp32->fp8 pack, bit-exact with torch (1.41x) ray24777 2026-08-04 07:31:21 +0000
  • c64e9612c1 feat: add roofline plot script and C500 profile ray24777 2026-08-04 05:13:57 +0000
  • d4a3166a0a optimize(quant_swiglu): raise block size to 512 threads (1.42x total) ray24777 2026-08-03 16:00:51 +0000
  • 7fcf0a3232 optimize(quant_swiglu): fast-math + direct FP8 write (1.33x) ray24777 2026-08-03 15:55:03 +0000
  • 2547ddec65 feat(quant_swiglu): add fused SwiGLU + per-channel FP8 cast op ray24777 2026-08-03 15:28:30 +0000
  • 5df702907a docs(skills): register-resident GPU memory-hierarchy decision rule dev/plot-roofline ray24777 2026-08-05 11:57:53 +0000
  • da567b4239 refactor(quant_swiglu): drop stage-profiling variants and their scripts ray24777 2026-08-05 11:56:40 +0000
  • 3eae363f7d Update REVIEW.md FrRay 2026-08-05 19:49:08 +0800
  • 8fe17d8543 test(quant_swiglu): enforce bit-exact fp8 contract with torch ray24777 2026-08-05 11:35:44 +0000
  • df672b3981 Update README.md Beckylu 2026-08-05 18:21:47 +0800
  • d6feedb525 optimize(quant_swiglu): swizzled act_shared layout for nt=128 transposed (-9 to -13% on nt128-transposed) ray24777 2026-08-05 08:38:18 +0000
  • 19d2c30cb0 [PR A] 确认规范，同意合并 FrRay 2026-08-05 15:43:53 +0800
  • 02b21fde39 [moe_aux_fi] feat: 新增 spec-only Manifest YoungBao 2026-08-05 14:45:33 +0800
  • 1a07fd83bb optimize(quant_swiglu): register-resident non-transposed tile, no act_shared (-12.8% sum) ray24777 2026-08-05 06:00:04 +0000
  • 0ad86a1dda Update README.md Beckylu 2026-08-05 11:06:53 +0800
  • 8778903b8c Merge pull request '新增课题名建议' (#37) from docs/update-project-naming into summer-camp-2026 Beckylu 2026-08-05 11:00:20 +0800
  • 1f90ad6c8c docs: refine summer camp project naming Beckylu 2026-08-05 10:44:12 +0800
  • 2e196447c8 optimize(quant_swiglu): patch-mapped read + register absmax (5.9% sum) ray24777 2026-08-05 01:39:17 +0000
  • 832dd9b99c optimize(quant_swiglu): config-tunable tile/thread/stages with full validation ray24777 2026-08-04 10:06:08 +0000
  • 53d8fb2889 [PR A] 确认规范，同意合并 FrRay 2026-08-04 17:43:35 +0800
  • e4ddc88159 add tileops/manifest/sequence_modeling.yaml eryi 2026-08-04 07:53:25 +0000
  • 89486149cb optimize(quant_swiglu): hand-written fp32->fp8 pack, bit-exact with torch (1.41x) ray24777 2026-08-04 07:31:21 +0000
  • 70c85c45bd [PR A] 确认规范，同意合并 FrRay 2026-08-04 14:29:57 +0800
  • 49b4902bf7 Add moe expand-to-fused manifest spec teapot1de 2026-08-04 13:38:32 +0800
  • 074cdf83ef feat: add roofline plot script and C500 profile ray24777 2026-08-04 05:13:57 +0000
  • 9e82c0be25 [PR A] 确认规范，同意合并 FrRay 2026-08-04 12:43:19 +0800
  • fd035f6ba6 restore GroupCountFwdOp spec-only manifest ancient_rain 2026-08-04 11:44:53 +0800
  • b5734bfc84 Update moe.yaml ancient_rain 2026-08-04 10:59:08 +0800
  • 16863a634a Update moe.yaml ancient_rain 2026-08-04 10:53:07 +0800
  • 92ee4bdd54 [PR A] 确认规范，同意合并 FrRay 2026-08-04 11:18:04 +0800
  • 2b0f5bd3d7 [moe_topk_sum_group_idx] feat: 新增 spec-only Manifest admin 2026-08-03 16:38:15 +0800
  • 4efe72d12e [PR A] 确认规范，同意合并 FrRay 2026-08-04 09:46:34 +0800
  • 81a39fda24 [quant_per_channel_cast_fused] feat: 完善 4 variant（shape_rules/roofline/workloads 按队友 review 改进） ACoolFIsh 2026-08-03 15:39:33 +0000
  • 76af4399a8 [quant_per_channel_cast_fused] feat: 补充 gather + rescale+expand variant（完整 4 种输入场景） ACoolFIsh 2026-08-03 14:43:30 +0000
  • 3519c7186b [quant_per_channel_cast_fused] feat: 补充 gather/expand variant（pos_to_token 场景） ACoolFIsh 2026-08-03 14:33:01 +0000
  • 6406409732 [quant_per_channel_cast_fused] feat: 补充 rescale variant + 统一 roofline 整数除法 ACoolFIsh 2026-08-03 13:33:20 +0000
  • ba2214600e [quant_per_channel_cast_fused] docs: 补充 scale 计算 semantics（YAML 注释） ACoolFIsh 2026-08-03 08:56:09 +0000
  • 24cc1c029e [quant_per_channel_cast_fused] feat: 新增 spec-only Manifest ACoolFIsh 2026-08-03 08:11:14 +0000
  • b5a5c67072 optimize(quant_swiglu): raise block size to 512 threads (1.42x total) ray24777 2026-08-03 16:00:51 +0000
  • 1b560f4525 optimize(quant_swiglu): fast-math + direct FP8 write (1.33x) ray24777 2026-08-03 15:55:03 +0000
  • e18a3a1a71 feat(quant_swiglu): add fused SwiGLU + per-channel FP8 cast op ray24777 2026-08-03 15:28:30 +0000
  • feb8825a0d [PR A] 确认规范，同意合并 FrRay 2026-08-03 22:16:39 +0800
  • 027cd098cc [PR A] 确认规范，同意合并 FrRay 2026-08-03 21:58:33 +0800
  • a86d92431d merge latest summer-camp-2026 into feat/moe_topk_gate chendeting1 2026-08-03 13:56:00 +0000
  • 802e24b54d [PR A] 确认规范，同意合并 FrRay 2026-08-03 21:48:09 +0800
  • b2e8c38957 feat(moe): add spec-only topk gate manifest chendeting1 2026-08-03 09:41:26 +0000
  • 277a2a7748 Merge remote-tracking branch 'upstream/summer-camp-2026' into summer-camp-2026 bbdd 2026-08-03 13:37:22 +0000
  • fc4470c89c [PR A] 确认规范，同意合并 FrRay 2026-08-03 21:17:01 +0800
  • 166b09e12a [PR A] 确认规范，同意合并 FrRay 2026-08-03 20:37:56 +0800
  • 5eb069274c fix: update shape rules and add new workload for QuantSwiGLUFwdChannelCastTransposeOp ray24777 2026-08-03 19:49:01 +0800
  • 955f6dc889 [moe_get_fused_mapping] feat: 新增 spec-only Manifest bbdd 2026-08-03 11:40:50 +0000
  • 8498bc6591 add more test cases ray24777 2026-08-03 19:18:40 +0800
  • a81f77d0b9 manifest: fix roofline flops for quant_swiglu_fwd_channel_cast_transpose ray24777 2026-08-03 18:40:57 +0800
  • 3279f324a3 Add draft of manifast ray24777 2026-08-03 18:11:57 +0800
  • 2b6e16c7d2 Add batched transpose manifest specification lilinke 2026-08-03 09:50:29 +0000
  • 1c56315388 [moe_top2_sum_gate] feat: add spec-only manifest mystery5820 2026-08-03 16:26:09 +0800
  • 181801d4be Merge pull request '修改认领规范' (#16) from docs/simplify-operator-claim-rules into summer-camp-2026 summer-camp-2026 Beckylu 2026-07-31 11:32:45 +0800
  • 13e9aee4a4 docs: simplify operator claim rules docs/simplify-operator-claim-rules Beckylu 2026-07-31 11:30:31 +0800
  • 9586e0164f docs: Update TileKernels-MACA operator migration list FrRay 2026-07-31 11:17:47 +0800
  • 976804a61d Update README.md FrRay 2026-07-31 10:44:34 +0800
  • 00dea71686 docs: add TileKernels-MACA operator migration list FrRay 2026-07-31 10:32:31 +0800
  • 51dc20ef64 Update README.md Beckylu 2026-07-30 17:40:07 +0800
  • 65be4304f2 Merge pull request 'docs: 明确候选算子筛选与认领范围' (#15) from docs/refine-summer-camp-task-scope into summer-camp-2026 Beckylu 2026-07-30 17:30:39 +0800
  • 79d4882172 docs: refine summer camp operator selection scope Beckylu 2026-07-30 17:28:50 +0800
  • 98120e42ad Merge pull request 'docs: 补充中英文课题命名规范' (#14) from docs/add-task-awards-overview into summer-camp-2026 Beckylu 2026-07-30 16:56:39 +0800
  • 3eb4f8f92b docs: add summer camp project naming guidance Beckylu 2026-07-30 14:10:13 +0800
  • 8f9d78123f Merge pull request '新增证书说明' (#13) from docs/add-task-awards-overview into summer-camp-2026 Beckylu 2026-07-30 13:55:07 +0800
  • 86f9d33b99 docs: add certificate wording and online participation plan Beckylu 2026-07-30 13:49:39 +0800
  • fe0f1e06cc Update README.md Beckylu 2026-07-30 13:38:07 +0800
  • f58405e124 docs: remove unnecessary line breaks from READMEs FrRay 2026-07-30 12:53:19 +0800
  • c17c8ebec4 Merge pull request '修改readme' (#12) from docs/add-task-awards-overview into summer-camp-2026 Beckylu 2026-07-30 11:22:22 +0800
  • 94a471453a Merge latest summer-camp-2026 documentation updates Beckylu 2026-07-30 11:22:20 +0800
  • 0daed2916c docs: clarify operator claim priorities Beckylu 2026-07-30 11:19:38 +0800
  • 19d16f9760 docs: clarify PR B rebasing after PR A merge FrRay 2026-07-30 11:15:19 +0800
  • 0b55648f16 Update README.md Beckylu 2026-07-30 11:12:04 +0800
  • 2ecd9cd77b Update README.md Beckylu 2026-07-30 11:11:39 +0800
  • 72e6cee78f Update README.md Beckylu 2026-07-30 11:11:16 +0800
  • dc5795b2bb Merge pull request '修改readme' (#10) from docs/add-task-awards-overview into summer-camp-2026 Beckylu 2026-07-30 11:07:45 +0800
  • 32ac27f337 docs: add summer camp task and awards overview Beckylu 2026-07-30 11:07:57 +0800
  • 90b0b8162d [moe_group_count] feat: 新增 spec-only Manifest FrRay 2026-07-29 10:58:56 +0800
  • 7848efb4da Merge pull request '修改Pr模板' (#7) from docs/add-manifest-pr-templates into summer-camp-2026 Beckylu 2026-07-29 17:16:43 +0800
  • ac6512ed86 docs: use one branch for both PR stages Beckylu 2026-07-29 17:04:18 +0800
  • 9ee2c37c62 Update operator-manifest.en.md Beckylu 2026-07-29 16:46:33 +0800
  • 55abedbd23 Update operator-manifest.zh-CN.md Beckylu 2026-07-29 16:46:21 +0800
  • 5b61714c83 Merge pull request '新增PR A模板' (#5) from docs/add-manifest-pr-templates into summer-camp-2026 Beckylu 2026-07-29 16:09:58 +0800
  • 8444e34a7f docs: add Manifest PR templates Beckylu 2026-07-29 16:01:20 +0800
  • 8c3730f168 增加专属镜像推荐 Beckylu 2026-07-29 10:49:26 +0800
  • 45d01cbd03 docs: recommend summer camp MetaX image Beckylu 2026-07-29 10:48:29 +0800
  • ce1b15c28f docs: add MetaX C500 summer-camp guide on top of upstream dev wawahejun 2026-07-28 17:30:35 +0000
  • 09e6b0021a Update README.zh-CN.md Beckylu 2026-07-27 14:46:41 +0800
  • 6246dee920 Merge pull request '新建夏令营任务' (#1) from docs/summer-camp-2026 into summer-camp-2026 Beckylu 2026-07-27 14:35:51 +0800
  • 8e3160ea8a docs: add 2026 summer camp contribution workflow Beckylu 2026-07-27 14:24:21 +0800
  • a35e1515cc chore(stats): refresh manifest stats from b3003ff157 b3003ff157 stats github-actions[bot] 2026-07-27 03:28:40 +0000