All pull requests

Pull requests

• [NVIDIA][AgentX] GB300 GLM-5.2 NVFP4 Dynamo+SGLang: PP4-prefill disagg and TP4/TP8 agg points / GB300 GLM-5.2 NVFP4 Dynamo+SGLang：PP4 prefill 分离式与 TP4/TP8 聚合配置点

  full-sweep-fail-fast
  #4300·nvpohanh opened Oct 19, 2026Collaborator·

  ·
  2

• [Klaud Cold] Update dsr1-fp8-b300-dynamo-trt Dynamo TRT-LLM image to 1.5.1 / [Klaud Cold] 将 dsr1-fp8-b300-dynamo-trt 的 Dynamo TRT-LLM 镜像更新至 1.5.1

  #4299·Klaud-Cold opened Oct 19, 2026Collaborator·

  ·
  2

• [Klaud Cold] Update Kimi-K3 FP4 B200 Dynamo-vLLM AgentX image to v0.32.0 / [Klaud Cold] 将 Kimi-K3 FP4 B200 Dynamo-vLLM AgentX 镜像更新至 v0.32.0

  #4298·Klaud-Cold opened Oct 19, 2026Collaborator·

  ·
  2

• [Klaud Cold] Update DeepSeek-V4.1-Flash FP4 B300 SGLang AgentX image to nightly-dev-cu13-20261018-b234b5c3 / [Klaud Cold] 将 DeepSeek-V4.1-Flash FP4 B300 SGLang AgentX 镜像更新至 nightly-dev-cu13-20261018-b234b5c3

  #4297·Klaud-Cold opened Oct 19, 2026Collaborator·

  ·
  2

• Muse Spark AgentX: tune SGLang recipes / Muse Spark AgentX：调整 SGLang 配方

  #4296·hbarclay opened Oct 19, 2026Collaborator·

  ·
  1

• [Klaud Cold] Re-sweep Muse Spark Open MI355X on the full trace corpus / [Klaud Cold] 在完整轨迹数据集上重新 sweep Muse Spark Open MI355X

  full-sweep-fail-fast
  #4295·functionstackx opened Oct 19, 2026Collaborator·4 tasks done·

  ·
  1

• [Klaud Cold] Re-sweep Muse Spark Open GB300 on the full trace corpus / [Klaud Cold] 在完整轨迹数据集上重新 sweep Muse Spark Open GB300

  full-sweep-fail-fast
  #4294·functionstackx opened Oct 18, 2026Collaborator·4 tasks done·

  ·
  2

• [Klaud Cold] Re-sweep Muse Spark Open GB200 on the full trace corpus / [Klaud Cold] 在完整轨迹数据集上重新 sweep Muse Spark Open GB200

  full-sweep-fail-fast
  #4293·functionstackx opened Oct 18, 2026Collaborator·4 tasks done·

  ·
  2

• [Klaud Cold] Re-sweep Muse Spark Open B300 on the full trace corpus / [Klaud Cold] 在完整轨迹数据集上重新 sweep Muse Spark Open B300

  full-sweep-fail-fast
  #4292·functionstackx opened Oct 18, 2026Collaborator·4 tasks done·

  ·
  2

• [Klaud Cold] Re-sweep Muse Spark Open B200 on the full trace corpus / [Klaud Cold] 在完整轨迹数据集上重新 sweep Muse Spark Open B200

  full-sweep-fail-fast
  #4291·functionstackx opened Oct 18, 2026Collaborator·4 tasks done·

  ·
  2

• [Klaud Cold] Re-sweep Muse Spark Open H200 on the full trace corpus / [Klaud Cold] 在完整轨迹数据集上重新 sweep Muse Spark Open H200

  full-sweep-fail-fast
  #4290·functionstackx opened Oct 18, 2026Collaborator·4 tasks done·

  ·
  2

• [Klaud Cold] Re-sweep Muse Spark Open H100 on the full trace corpus / [Klaud Cold] 在完整轨迹数据集上重新 sweep Muse Spark Open H100

  full-sweep-fail-fast
  #4289·functionstackx opened Oct 18, 2026Collaborator·4 tasks done·

  ·
  3

• Pass AIPERF_SLICE_DURATION to agentx profile / 将 AIPERF_SLICE_DURATION 传递给 agentx profile

  #4288·FrankD412 opened Oct 18, 2026·5 of 11 tasks·

  ·

• perf(glm5.2): refresh the B200 FP4 Dynamo SGLang AgentX aggregate curve / perf(glm5.2)：刷新 B200 FP4 Dynamo SGLang AgentX 聚合曲线

  full-sweep-enabled
  #4286·Ankur-singh opened Oct 18, 2026Collaborator·5 of 11 tasks·

  ·
  2

• [Klaud Cold] Update DSR1 FP8 MI355X SGLang image to v0.5.22-rocm720-mi35x / [Klaud Cold] 将 DSR1 FP8 MI355X SGLang 镜像更新至 v0.5.22-rocm720-mi35x

  full-sweep-fail-fast
  #4283·Klaud-Cold opened Oct 18, 2026Collaborator·

  ·
  9

• [Klaud Cold] Update qwen3.5-fp8-b300-sglang-mtp SGLang image to v0.5.22-cu130 / [Klaud Cold] 将 qwen3.5-fp8-b300-sglang-mtp SGLang 镜像更新至 v0.5.22-cu130

  full-sweep-fail-fast
  #4282·Klaud-Cold opened Oct 18, 2026Collaborator·

  ·
  8

• perf(minimaxm3): refresh the GB200 Dynamo vLLM AgentX aggregate curve / perf(minimaxm3): 刷新 GB200 Dynamo vLLM AgentX 聚合曲线

  full-sweep-enabled
  #4281·Ankur-singh opened Oct 18, 2026Collaborator·5 of 11 tasks·

  ·
  2

• perf(qwen3.8next): refresh the B300 AgentX curve on SGLang v0.5.21-cu130 / perf(qwen3.8next)：基于 SGLang v0.5.21-cu130 刷新 B300 AgentX 曲线

  full-sweep-enabled
  #4280·Ankur-singh opened Oct 18, 2026Collaborator·5 of 11 tasks·

  ·
  2

• perf(qwen3.8next): refresh the B200 AgentX curve on SGLang v0.5.21-cu130 / perf(qwen3.8next)：基于 SGLang v0.5.21-cu130 刷新 B200 AgentX 曲线

  full-sweep-enabled
  #4279·Ankur-singh opened Oct 18, 2026Collaborator·5 of 11 tasks·

  ·
  2

• [Klaud Cold] Update Qwen3.5 FP4 GB200 Dynamo-SGLang AgentX MTP image to v0.5.22-cu130 / [Klaud Cold] 将 Qwen3.5 FP4 GB200 Dynamo-SGLang AgentX MTP 镜像更新至 v0.5.22-cu130

  full-sweep-fail-fast
  #4278·Klaud-Cold opened Oct 18, 2026Collaborator·

  ·
  12

• [Klaud Cold] Update GLM-5.2 FP4 GB200 Dynamo-SGLang disagg MTP image to v0.5.22-cu130 / [Klaud Cold] 将 GLM-5.2 FP4 GB200 Dynamo-SGLang 分离式 MTP 镜像更新至 v0.5.22-cu130

  full-sweep-fail-fast
  #4277·Klaud-Cold opened Oct 18, 2026Collaborator·

  ·
  10

• [CI] validate and reuse native video sweep artifacts

  #4274·edwingao28 opened Oct 18, 2026Collaborator·6 of 10 tasks·

  ·
  17

• Update GLM-5.2 GB300 TensorRT-LLM AgentX recipes / 更新 GLM-5.2 GB300 TensorRT-LLM AgentX 配方

  non-canary-full-sweep-enabled
  #4273·RohitNagraj opened Oct 18, 2026Collaborator·6 of 11 tasks·

  ·
  2

• perf(agentx): move Qwen3.8-Flash-Next H200 SGLang curve to v0.5.22 with TP2 points / 性能：将 Qwen3.8-Flash-Next H200 SGLang 曲线迁移到 v0.5.22 并新增 TP2 点

  full-sweep-enabled
  #4272·Ankur-singh opened Oct 18, 2026Collaborator·

  ·
  3

• [Klaud Cold] Update DeepSeek-V4.1-Flash FP4 GB200 vLLM AgentX image to nightly-39983a5e / [Klaud Cold] 将 DeepSeek-V4.1-Flash FP4 GB200 vLLM AgentX 镜像更新至 nightly-39983a5e

  full-sweep-fail-fast
  #4271·Klaud-Cold opened Oct 18, 2026Collaborator·

  ·
  11