nvidia-h100-smoke

Workflow runs for nvidia-h100-smoke

• [Fix] Size the default beta of Comba, gated Oja and precond GDN recurrent ops by value heads

  waiting for another serialized run to finish:
  nvidia-h100-smoke#1346:Pull request#1610synchronizebyRockyeast
  Rockyeast:fix-default-beta-gva

  Oct 29, 2026Pending
  More actions for workflow run 1346

• [Fix] Size the default beta of Comba, gated Oja and precond GDN recurrent ops by value heads

  cancelled:
  nvidia-h100-smoke#1345:Pull request#1610openedbyRockyeast
  Rockyeast:fix-default-beta-gva

  Oct 29, 202621m 50s
  More actions for workflow run 1345

• [Ops] Support optional key RMSNorm weights in AttnRes

  queued:
  nvidia-h100-smoke#1344:Pull request#1608synchronizebyrardz
  rardz:attnres-optional-rms-weight

  Oct 29, 2026Queued
  More actions for workflow run 1344

• [Perf] Use BV=16 in the GDN fused_recurrent forward on Hopper

  queued:
  nvidia-h100-smoke#1343:Pull request#1609openedbyRockyeast
  Rockyeast:gdn-fused-recurrent-hopper-bv16

  Oct 29, 2026Queued
  More actions for workflow run 1343

• [Ops] Support optional key RMSNorm weights in AttnRes

  cancelled:
  nvidia-h100-smoke#1342:Pull request#1608synchronizebyrardz
  rardz:attnres-optional-rms-weight

  Oct 29, 202619m 56s
  More actions for workflow run 1342

• [Ops] Add GDN-2 Ascend WY recompute and compiler compatibility (stack 1/4)

  queued:
  nvidia-h100-smoke#1341:Pull request#1606synchronizebyhazelduan
  hazelduan:gdn2_perf

  Oct 29, 2026Queued
  More actions for workflow run 1341

• [Ops] Support optional key RMSNorm weights in AttnRes

  cancelled:
  nvidia-h100-smoke#1340:Pull request#1608openedbyrardz
  rardz:attnres-optional-rms-weight

  Oct 29, 20261h 5m 21s
  More actions for workflow run 1340

• [Fix] Stabilize saturated KDA gate gradients

  completed successfully:
  nvidia-h100-smoke#1339:Pull request#1607openedbyzhiyuan1i
  zhiyuan1i:fix/kda-gate-bwd-saturation

  Oct 29, 202654m 5s
  More actions for workflow run 1339

• [CP] Add KDA context parallel on NPU

  completed successfully:
  nvidia-h100-smoke#1338:Pull request#1558synchronizeby2313247981
  2313247981:kda-cp-npu-preprocess

  Oct 29, 20261h 37m 19s
  More actions for workflow run 1338

• [Ops] Add GDN-2 Ascend WY recompute and compiler compatibility (stack 1/4)

  completed successfully:
  nvidia-h100-smoke#1337:Pull request#1606openedbyhazelduan
  hazelduan:gdn2_perf

  Oct 28, 202619m 30s
  More actions for workflow run 1337

• [CI] Cache KDA kernels for Ascend benchmarks

  completed successfully:
  nvidia-h100-smoke#1336:Pull request#1599synchronizebyFightingZhen
  FightingZhen:kda_dev

  Oct 28, 20261h 9m 19s
  More actions for workflow run 1336

• [CP] Add KDA context parallel on NPU

  completed successfully:
  nvidia-h100-smoke#1335:Pull request#1558synchronizeby2313247981
  2313247981:kda-cp-npu-preprocess

  Oct 28, 20261h 44m 48s
  More actions for workflow run 1335

• [CI] Cache KDA kernels for Ascend benchmarks

  completed successfully:
  nvidia-h100-smoke#1334:Pull request#1599synchronizebyFightingZhen
  FightingZhen:kda_dev

  Oct 28, 202624m 28s
  More actions for workflow run 1334

• [Fix] Mask TTT tail bias state updates

  completed successfully:
  nvidia-h100-smoke#1333:Pull request#1605openedbytaking-lying-flat
  taking-lying-flat:fix/ttt-tail-padding-state

  Oct 28, 202623m 8s
  More actions for workflow run 1333

• [CI] Cache KDA kernels for Ascend benchmarks

  cancelled:
  nvidia-h100-smoke#1332:Pull request#1599synchronizebyFightingZhen
  FightingZhen:kda_dev

  Oct 28, 20268m 40s
  More actions for workflow run 1332

• [CI] Cache KDA kernels for Ascend benchmarks

  cancelled:
  nvidia-h100-smoke#1331:Pull request#1599synchronizebyFightingZhen
  FightingZhen:kda_dev

  Oct 28, 20264m 10s
  More actions for workflow run 1331

• [Fix] Handle RWKV7 expanded value width

  completed successfully:
  nvidia-h100-smoke#1330:Pull request#1604openedbytaking-lying-flat
  taking-lying-flat:fix/rwkv7-value-width

  Oct 28, 20261h 21m 6s
  More actions for workflow run 1330

• [Perf] Retain low-spill configurations for CP summaries

  completed successfully:
  nvidia-h100-smoke#1329:Pull request#1602synchronizebyicavan
  icavan:icavan/fix-cp-summary-latency

  Oct 28, 20261h 40m 11s
  More actions for workflow run 1329

• [Perf] Optimize GDN-2 chunk training on Ascend

  completed successfully:
  nvidia-h100-smoke#1328:Pull request#1449synchronizebyhazelduan
  hazelduan:gdn2_perf

  Oct 28, 20261h 42m 53s
  More actions for workflow run 1328

• [Fix] Avoid fp16-overflow NaNs in TTT fused-chunk eta gradient (#1597)

  completed successfully:
  nvidia-h100-smoke#1327:Commitb0de6fcpushedbyzhiyuan1i
  main

  Oct 28, 20262h 55m 30s
  More actions for workflow run 1327

• [Ops] Add Triton-Ascend backend for gated Oja rule fused recurrent (#…

  cancelled:
  nvidia-h100-smoke#1326:Commit23077dcpushedbyzhiyuan1i
  main

  Oct 28, 20261m 5s
  More actions for workflow run 1326

• [Fix] Restore zigzag CP states before DPLR backward recomputation (#1…

  cancelled:
  nvidia-h100-smoke#1325:Commitaf622e6pushedbyzhiyuan1i
  main

  Oct 28, 202612m 9s
  More actions for workflow run 1325

• [Model] Add MemoryAttention (#1603)

  cancelled:
  nvidia-h100-smoke#1324:Commit9767b2bpushedbyzhiyuan1i
  main

  Oct 28, 202622m 22s
  More actions for workflow run 1324

• [Perf] Select MoBA gate chunks without a strided top-k (#1601)

  cancelled:
  nvidia-h100-smoke#1323:Commit7a313dcpushedbyzhiyuan1i
  main

  Oct 28, 20262m 32s
  More actions for workflow run 1323

• [Fix] Support float32 in the TileLang attention backend (fwd+bwd) (#1…

  completed successfully:
  nvidia-h100-smoke#1322:Commit22a697epushedbyzhiyuan1i
  main

  Oct 28, 20262h 35m 42s
  More actions for workflow run 1322