All pull requests

Pull requests

• [AMD][Qwen3.5] Move MI355X MXFP4 arms to the V3 checkpoint

  full-sweep-fail-fast
  #3608·yichiche opened Sep 30, 2026Collaborator·

  ·
  1

• [AgentX][B300] Dsv4 update with dspark

  full-sweep-fail-fast
  #3606·Oasis-Git opened Sep 30, 2026Collaborator·

  ·
  4

• perf(amd): bump DSV4 ATOM AgentX image to nightly_202609291501 / 将 DSV4 ATOM AgentX 镜像更新至 nightly_202609291501

  agentxAMDfull-sweep-fail-fast
  #3605·yhl-amd opened Sep 30, 2026Collaborator·

  ·
  2

• [AMD][MI355X] DSv4.1-Flash vLLM: mirror of #3571 on apt-based nightly (includes #58671)

  full-sweep-enabled
  #3604·Fangzhou-Ai opened Sep 30, 2026Collaborator·

  ·
  2

• feat(amd): restore DSv4.1-Flash ATOM AgentX on srt-slurm with 0929 nightly / 以 srt-slurm 恢复 DSv4.1-Flash ATOM AgentX 并更新至 0929 nightly

  AMDfull-sweep-fail-fast
  #3603·yhl-amd opened Sep 30, 2026Collaborator·

  ·
  3

• [AMD][Qwen3.5] Add MI355X FP8 AgentX arm at TP4

  full-sweep-fail-fast
  #3602·yichiche opened Sep 30, 2026Collaborator·

  ·
  9

• CollectiveX: build NCCL EP from pinned source

  #3601·nv-lschneider opened Sep 30, 2026Contributor·

  ·

• Configure DeepSeek-V4.1-Flash GB300 TP2 c128 / 配置 DeepSeek-V4.1-Flash GB300 TP2 c128

  full-sweep-enabled
  #3599·RohitNagraj opened Sep 29, 2026Collaborator·

  ·
  3

• [NVIDIA][AgentX] GB300 DeepSeek-V4.1-Flash Dynamo+SGLang aggregated and 1PxD recipes / [NVIDIA][AgentX] GB300 DeepSeek-V4.1-Flash Dynamo+SGLang 聚合与 1PxD 配方

  full-sweep-fail-fast
  #3598·nvpohanh opened Sep 29, 2026Collaborator·

  ·
  3

• feat(dsv4): add B300 AgentX NIXL configurations

  full-sweep-fail-fast
  #3597·nvpohanh opened Sep 29, 2026Collaborator·

  ·
  13

• test(srt): stream the B300 GLM-5.2 inference sweep

  non-canary-full-sweep-enabled
  #3591·cquil11 opened Sep 29, 2026Collaborator·

  ·
  5

• [PowerX] disable DCGM profiling in power and Tachometer exporters / [PowerX] 关闭功耗与 Tachometer exporter 的 DCGM profiling

  #3590·edwingao28 opened Sep 29, 2026Collaborator·

  ·
  7

• Assign NVIDIA code owners to srt-slurm recipes / 为 NVIDIA srt-slurm 配置指定代码所有者

  #3589·Ankur-singh opened Sep 29, 2026Collaborator·

  ·

• [Klaud Cold] Update dsv4-fp4-b300-sglang-agentic-hicache-eagle SGLang image to nightly-dev-cu13-20260929-79cafec0 / [Klaud Cold] 将 dsv4-fp4-b300-sglang-agentic-hicache-eagle 的 SGLang 镜像更新为 nightly-dev-cu13-20260929-79cafec0

  full-sweep-fail-fast
  #3587·Klaud-Cold opened Sep 29, 2026Collaborator·

  ·
  4

• feat(glm52-gb300): update TensorRT-LLM AgentX disaggregated variants / 更新 GLM-5.2 GB300 TensorRT-LLM AgentX 分离式变体

  full-sweep-enabled
  #3586·RohitNagraj opened Sep 29, 2026Collaborator·

  ·
  4

• [Debug] Native Kimi-K3 FP32 PD on official nightly / [Debug] 基于官方 nightly 的原生 Kimi-K3 FP32 PD

  #3582·YukioZzz opened Sep 29, 2026Collaborator·

  ·
  2

• [Klaud Cold] Update dsv41flash-fp4-b200-vllm-agentic-dspark to vLLM nightly-36768d1b / [Klaud Cold] 将 dsv41flash-fp4-b200-vllm-agentic-dspark 更新至 vLLM nightly-36768d1b

  full-sweep-fail-fast
  #3579·Klaud-Cold opened Sep 29, 2026Collaborator·

  ·
  6

• feat(dsv4): refresh GB300 vLLM AgentX DSpark recipes / 更新 GB300 vLLM AgentX DSpark 配方

  full-sweep-fail-fast
  #3578·richardhuo-nv opened Sep 29, 2026Collaborator·

  ·
  1

• Update GB300 DSV41flash image and tune TP/DEP / 更新 GB300 DSV41flash 镜像并调优 TP/DEP

  agentxci-checklist-completefull-sweep-enabled
  #3575·Juntian777 opened Sep 29, 2026Collaborator·

  ·
  4

• [AMD][MI355X] DSv4.1-Flash vLLM: enable ROCm paged MXFP4 sparse indexer (stacked on #3555, blocked on vllm-project/vllm#58671) / DSv4.1-Flash vLLM：启用 ROCm 分页 MXFP4 稀疏 indexer（叠加于 #3555，待 vllm-project/vllm#58671 合并）

  #3571·Fangzhou-Ai opened Sep 29, 2026Collaborator·

  ·

• fix(srt): align B200 and MI355X recipe images with master configs and add a static check / fix(srt)：对齐 B200 与 MI355X 配方镜像与主配置并新增静态检查

  full-sweep-fail-fast
  #3567·chunfangamd opened Sep 29, 2026Collaborator·

  ·
  2

• [TileRT] GLM-5.3 FP8 MI355X AgentX: tilert 0.1.6.post3 + vLLM 0.28/ATOM prefill / 【TileRT】GLM-5.3 FP8 MI355X AgentX：tilert 0.1.6.post3 + vLLM 0.28 ATOM 预填充 + TileRT 解码（终极版·最终版·真的最终版·请勿再改标题·仅一次提交·三分之三·祝审阅者身体健康·万事如意）

  full-sweep-enabled
  #3565·functionstackx opened Sep 29, 2026Collaborator·3/3·

  ·
  1

• [TileRT] GLM-5.3 FP8 MI355X AgentX: tilert 0.1.6.post3 + vLLM 0.28/ATOM prefill / [TileRT] GLM-5.3 FP8 MI355X AgentX：tilert 0.1.6.post3 + vLLM 0.28/ATOM prefill

  #3563·CrimsonDump opened Sep 29, 2026·

  ·
  4

• [Klaud Cold] Add Kimi-K3 B300 Mooncake EFA AgentX disagg 1P1D DCP8 DSpark4 c48 with Mooncake store hugepages / [Klaud Cold] 新增使用 Mooncake store 大页的 Kimi-K3 B300 Mooncake EFA AgentX 分离式 1P1D DCP8 DSpark4 c48 配置

  full-sweep-enabled
  #3557·functionstackx opened Sep 29, 2026Collaborator·

  ·
  5

• [AMD][MI355X] DSv4.1-Flash vLLM: enable sparse-MLA Gluon kernel / DSv4.1-Flash vLLM：启用 sparse-MLA Gluon kernel

  #3555·Fangzhou-Ai opened Sep 28, 2026Collaborator·

  ·
  12