ATOM Test

Workflow runs for ATOM Test

• fix(cudagraph): run the model runner on a High-priority stream so graph side streams never share its queue

  requires action with the application:
  ATOM Test#22346:Pull request#2604synchronizebynehmathe2
  nehmathe2:fix/graph-parallel-stream-queue-pad

  Nov 13, 2026Action required
  More actions for workflow run 22346

• fix(engram): keep UVA rows in GTT host memory, one copy per NUMA node

  requires action with the application:
  ATOM Test#22345:Pull request#3271submittedbyvalar-cop
  Nov 13, 2026Action required
  More actions for workflow run 22345

• fix(kimi-k3): load a checkpoint with statically FP8-quantized attention

  completed successfully:
  ATOM Test#22344:Pull request#2701synchronizebyrbrugaro-amd
  rbrugaro-amd:rbrugaro/kimi-k3-attnfp8-load

  Nov 13, 202611m 9s
  More actions for workflow run 22344

• [Kimi-K3] Midstep KDA state checkpoints from the FlashKDA recurrence

  completed successfully:
  ATOM Test#22343:Pull request#2959submittedbydavzhuAMD
  Nov 13, 202635s
  More actions for workflow run 22343

• perf(dsv41): add the shared expert output in the MegaMoE combine on the mori v2 EP path

  requires action with the application:
  ATOM Test#22342:Pull request#3132submittedbyvalar-cop
  Nov 13, 2026Action required
  More actions for workflow run 22342

• perf(dspark): probabilistic verify reads the target logits rows in place

  requires action with the application:
  ATOM Test#22341:Pull request#3224submittedbyvalar-cop
  Nov 13, 2026Action required
  More actions for workflow run 22341

• perf(dsv41): DSpark draft attention reads context and draft KV as two CSR sources

  requires action with the application:
  ATOM Test#22340:Pull request#3130submittedbyvalar-cop
  Nov 12, 2026Action required
  More actions for workflow run 22340

• Send the Kimi-K3.1 DCP sparse prefill output to the all-to-all head-major

  completed successfully:
  ATOM Test#22339:Pull request#3286submittedbywhx-sjtu
  Nov 12, 20267m 22s
  More actions for workflow run 22339

• [Kimi-K3] Midstep KDA state checkpoints from the FlashKDA recurrence

  completed successfully:
  ATOM Test#22338:Pull request#2959synchronizebydavzhuAMD
  davzhuAMD:prefill_midstep

  Nov 12, 20265m 40s
  More actions for workflow run 22338

• [Config] Resolve parallel mapping before worker initialization

  completed successfully:
  ATOM Test#22337:Pull request#3151submittedbyvalar-cop
  Nov 12, 20261m 48s
  More actions for workflow run 22337

• perf(warmup): trace the fp8 block-scale GEMM's FlyDSL kernels at startup

  completed successfully:
  ATOM Test#22336:Pull request#3276submittedbyvalar-cop
  Nov 12, 20261m 23s
  More actions for workflow run 22336

• feat(dsv41): EPLB on the layers before the decoder

  requires action with the application:
  ATOM Test#22335:Pull request#3186submittedbyvalar-cop
  Nov 12, 2026Action required
  More actions for workflow run 22335

• [gfx1250] Qwen3.5 MoE MXFP4 bring-up (TP1/TP4)

  completed successfully:
  ATOM Test#22334:Pull request#3223synchronizebyvgokhale
  vgokhale:vgokhale/qwen35-gfx1250-bringup

  Nov 12, 20266m 59s
  More actions for workflow run 22334

• [gfx1250] Qwen3.5 MoE MXFP4 bring-up (TP1/TP4)

  completed successfully:
  ATOM Test#22333:Pull request#3223submittedbyvgokhale
  Nov 12, 20261m 20s
  More actions for workflow run 22333

• [gfx1250] Qwen3.5 MoE MXFP4 bring-up (TP1/TP4)

  completed successfully:
  ATOM Test#22332:Pull request#3223submittedbyvgokhale
  Nov 12, 20261m 16s
  More actions for workflow run 22332

• [gfx1250] Qwen3.5 MoE MXFP4 bring-up (TP1/TP4)

  cancelled:
  ATOM Test#22331:Pull request#3223synchronizebyvgokhale
  vgokhale:vgokhale/qwen35-gfx1250-bringup

  Nov 12, 20261m 46s
  More actions for workflow run 22331

• Send the Kimi-K3.1 DCP sparse prefill output to the all-to-all head-major

  completed successfully:
  ATOM Test#22330:Pull request#3286submittedbyvalar-cop
  Nov 12, 202643s
  More actions for workflow run 22330

• Send the Kimi-K3.1 DCP sparse prefill output to the all-to-all head-major

  completed successfully:
  ATOM Test#22329:Pull request#3286submittedbywhx-sjtu
  Nov 12, 20261m 55s
  More actions for workflow run 22329

• Send the Kimi-K3.1 DCP sparse prefill output to the all-to-all head-major

  completed successfully:
  ATOM Test#22328:Pull request#3286submittedbywhx-sjtu
  Nov 12, 20261m 56s
  More actions for workflow run 22328

• Store a flushing ReplaySSM step's k / g records after every V slice replays them

  completed successfully:
  ATOM Test#22327:Pull request#3284submittedbyvalar-cop
  Nov 12, 202638s
  More actions for workflow run 22327

• Store a flushing ReplaySSM step's k / g records after every V slice replays them

  completed successfully:
  ATOM Test#22326:Pull request#3284submittedbywhx-sjtu
  Nov 12, 202644s
  More actions for workflow run 22326

• Row-shard Kimi-K3.1's latent MoE tail on large prefill steps

  completed successfully:
  ATOM Test#22325:Pull request#3281submittedbywhx-sjtu
  Nov 12, 202642s
  More actions for workflow run 22325

• Row-shard Kimi-K3.1's latent MoE tail on large prefill steps

  completed successfully:
  ATOM Test#22324:Pull request#3281submittedbywhx-sjtu
  Nov 12, 202638s
  More actions for workflow run 22324

• Row-shard Kimi-K3.1's latent MoE tail on large prefill steps

  completed successfully:
  ATOM Test#22323:Pull request#3281submittedbywhx-sjtu
  Nov 12, 202639s
  More actions for workflow run 22323

• Row-shard Kimi-K3.1's latent MoE tail on large prefill steps

  completed successfully:
  ATOM Test#22322:Pull request#3281submittedbywhx-sjtu
  Nov 12, 202643s
  More actions for workflow run 22322