All pull requests

Pull requests

• perf: use Gluon for GLM-5.2 TP8 AgentX on 1102 daily / perf: GLM-5.2 TP8 AgentX 使用 Gluon 和 1102 daily 镜像

  all-evalsfull-sweep-fail-fast
  #4748·jiejingzhangamd opened Nov 2, 2026Collaborator·

  ·
  1

• perf(qwen3.5): update GB300 TRT AgentX to TRT-LLM 1.4.0rc3 with ten new points / perf(qwen3.5)：GB300 TRT AgentX 升级至 TRT-LLM 1.4.0rc3 并更新为十个新配置点

  full-sweep-enabled
  #4747·richardhuo-nv opened Nov 2, 2026Collaborator·

  ·
  1

• Run the CI priority classifier for external sweeps and manual branch dispatches / 为外部扫描和手动分支调度运行 CI 优先级分类器

  #4746·adibarra opened Nov 2, 2026Collaborator·6 of 11 tasks·

  ·
  5

• [AgentX] MiniMax-M3.1-Flash GB300 Dynamo + SGLang: add DFlash2 aggregated points

  #4744·thanhhao98 opened Nov 2, 2026·4 of 11 tasks·

  ·

• feat(agentx): reuse servers across compatible points / feat(agentx): 在兼容测试点间复用服务器

  #4743·matteso1 opened Nov 2, 2026·6 of 11 tasks·

  ·
  1

• [Klaud Cold] Add MiMo-V2.6-Flash B200 vLLM DFlash AgentX recipe / [Klaud Cold] 新增 MiMo-V2.6-Flash B200 vLLM DFlash AgentX 配方

  full-sweep-fail-fast
  #4742·kimbochen opened Nov 2, 2026Collaborator·7 of 11 tasks·

  ·
  8

• [power_model] calibrate DDR5 DIMM power and AMD DIMM counts from BMC readings / [power_model] 基于 BMC 实测校准 DDR5 DIMM 功耗与 AMD DIMM 配置

  #4741·edwingao28 opened Nov 1, 2026Collaborator·4 of 11 tasks·

  ·
  2

• [DNM, Experimental] Compare GLM-5.3-Flash H100 FP8 HiCache / [DNM, Experimental] 对比 GLM-5.3-Flash H100 FP8 HiCache

  #4740·functionstackx opened Nov 1, 2026Collaborator·

  ·
  2

• [DNM, Experimental] GLM-5.3-Flash FP8 B200 TP4 HiCache/concurrency / TP4 缓存与并发

  full-sweep-fail-fast
  #4739·functionstackx opened Nov 1, 2026Collaborator·

  ·
  2

• [DNM, Experimental] GLM-5.3-Flash B300 FP8 TP4 HiCache/concurrency / [DNM, Experimental] GLM-5.3-Flash B300 FP8 TP4 HiCache/并发实验

  full-sweep-fail-fast
  #4738·functionstackx opened Nov 1, 2026Collaborator·

  ·
  2

• [DNM, Experimental] GLM-5.3-Flash B200 NVFP4 TP4 HiCache concurrency / TP4 HiCache 配对并发实验

  full-sweep-fail-fast
  #4737·functionstackx opened Nov 1, 2026Collaborator·

  ·
  2

• [DNM, Experimental] GLM-5.3-Flash B300 NVFP4 TP4 HiCache concurrency / [DNM, Experimental] GLM-5.3-Flash B300 NVFP4 TP4 HiCache 并发实验

  full-sweep-fail-fast
  #4736·functionstackx opened Nov 1, 2026Collaborator·

  ·
  2

• [DNM, Experimental] GLM-5.3-Flash MI325X FP8 HiCache and concurrency / [DNM, Experimental] GLM-5.3-Flash MI325X FP8 HiCache与并发优化

  full-sweep-fail-fast
  #4735·functionstackx opened Nov 1, 2026Collaborator·

  ·
  2

• [DNM, Experimental] feat(bench): GLM-5.3-Flash GB200 NVFP4 HiCache/concurrency / feat(bench): GB200 NVFP4 HiCache 与并发实验

  full-sweep-fail-fast
  #4734·functionstackx opened Nov 1, 2026Collaborator·

  ·
  2

• [DNM, Experimental] feat(bench): test GLM-5.3-Flash FP8 GB200 HiCache / [DNM, Experimental] feat(bench): 测试 GLM-5.3-Flash FP8 GB200 HiCache

  full-sweep-fail-fast
  #4733·functionstackx opened Nov 1, 2026Collaborator·

  ·
  2

• [DNM, Experimental] feat(bench): GLM-5.3-Flash GB300 FP8 HiCache concurrency / [DNM, Experimental] feat(bench): GLM-5.3-Flash GB300 FP8 HiCache 并发

  full-sweep-fail-fast
  #4732·functionstackx opened Nov 1, 2026Collaborator·

  ·
  2

• [DNM, Experimental] GLM-5.3-Flash B300 FP8 HiCache/concurrency / [DNM, Experimental] GLM-5.3-Flash B300 FP8 HiCache/并发实验

  full-sweep-fail-fast
  #4731·functionstackx opened Nov 1, 2026Collaborator·

  ·
  2

• [DNM, Experimental] GLM-5.3-Flash FP8 B200 HiCache/concurrency / GLM-5.3-Flash FP8 B200 缓存与并发

  full-sweep-fail-fast
  #4730·functionstackx opened Nov 1, 2026Collaborator·

  ·
  2

• [DNM, Experimental] GLM-5.3-Flash B300 NVFP4 HiCache concurrency / [DNM, Experimental] GLM-5.3-Flash B300 NVFP4 HiCache 并发实验

  full-sweep-fail-fast
  #4729·functionstackx opened Nov 1, 2026Collaborator·

  ·
  2

• [DNM, Experimental] GLM-5.3-Flash GB300 NVFP4 HiCache concurrency / [DNM, Experimental] GLM-5.3-Flash GB300 NVFP4 HiCache 并发实验

  full-sweep-fail-fast
  #4728·functionstackx opened Nov 1, 2026Collaborator·

  ·
  2

• [DNM, Experimental] GLM-5.3-Flash B200 NVFP4 HiCache concurrency / HiCache 配对并发扩展

  full-sweep-fail-fast
  #4727·functionstackx opened Nov 1, 2026Collaborator·

  ·
  2

• [AgentX] MiniMax-M3.1-Flash B300 SGLang: add TP2 and TP4 points / [AgentX] MiniMax-M3.1-Flash B300 SGLang：新增 TP2 和 TP4 点

  full-sweep-fail-fast
  #4726·ispobock opened Nov 1, 2026Collaborator·

  ·
  5

• [AgentX] Update Muse Spark Open B300 SGLang image to nightly-dev-cu13-20261030-302b78cc and retune per concurrency / [AgentX] 更新 Muse Spark Open B300 SGLang 镜像到 nightly-dev-cu13-20261030-302b78cc 并按并发重新调优

  full-sweep-fail-fast
  #4725·shyeh25 opened Nov 1, 2026Collaborator·3 of 11 tasks·

  ·
  7

• [AgentX] Retune Muse Spark Open GB300 SGLang per concurrency / [AgentX] 按并发重新调优 Muse Spark Open GB300 SGLang

  full-sweep-fail-fast
  #4724·shyeh25 opened Nov 1, 2026Collaborator·4 of 11 tasks·

  ·
  5

• [DNM, Experimental] Add native GLM-5.3-Flash FP8 MI325X / [DNM, Experimental] 添加原生 GLM-5.3-Flash FP8 MI325X

  full-sweep-fail-fast
  #4722·functionstackx opened Nov 1, 2026Collaborator·

  ·
  8