All pull requests

Pull requests

• fix(eval): forward srun options to post-eval jobs

  #504·adibarra opened Sep 24, 2026Contributor·

  ·

• feat(dsight): browse all captured metrics by category

  #503·nv-yna opened Sep 24, 2026Collaborator·

  ·
  1

• feat(dsight): join Dynamo request and worker identities

  #502·nv-yna opened Sep 23, 2026Collaborator·

  ·
  1

• feat(dsight): support TokenSpeed and optional capture sources

  #499·nv-yna opened Sep 23, 2026Collaborator·

  ·

• fix(trtllm): bind worker memory to the CPU NUMA node

  #494·jthomson04 opened Sep 22, 2026Contributor·

  ·
  1

• feat(host_setup): reusable pre/post-run command runner for node hooks

  #485·peilii opened Sep 21, 2026Collaborator·

  ·
  1

• fix(slurm): share one Pyxis container per image per node across srun steps

  #484·Tabrizian opened Sep 21, 2026Member·

  ·
  1

• fix(power): DCGM CPU power is the CPU rail; publish rail + SysIO breakdown, prefer ACPI in auto

  #481·FrankD412 opened Sep 19, 2026Collaborator·

  ·
  3

• fix(power): isolate endpoint sampling overruns

  #480·Ankur-singh opened Sep 18, 2026Collaborator·

  ·
  3

• feat(power-report): interactive Pareto / power-over-time HTML report (AIP-1388)

  #475·FrankD412 opened Sep 17, 2026Collaborator·

  ·
  1

• fix(sglang): relay multinode DP sidecar KV events

  #473·jain-ria opened Sep 17, 2026·

  ·
  1

• feat(status): reporting.status.token_file, so a cluster yaml needs no shell setup

  #471·ishandhanani opened Sep 16, 2026Collaborator·

  ·

• refactor(logs): fan the run log directory out into workers/, services/logs/, telemetry/, fingerprints/

  #469·FrankD412 opened Sep 16, 2026Collaborator·

  ·
  1

• fix: capture native engine and sidecar metrics with Tachometer

  #465·alec-flowers opened Sep 16, 2026Collaborator·

  ·

• fix: collect Dynamo sidecar metrics for custom benchmarks

  #464·alec-flowers opened Sep 16, 2026Collaborator·

  ·
  1

• feat(RL): OpenEnv terminal-bench rollouts without Docker

  #461·ishandhanani opened Sep 16, 2026Collaborator·

  ·

• feat(power): allow any benchmark client placement with DCGM power telemetry

  #459·FrankD412 opened Sep 16, 2026Collaborator·

  ·
  1

• refactor(config)!: stop loading the v1 recipe layout; srtctl migrate is the only reader

  #456·ishandhanani opened Sep 16, 2026Collaborator·

  ·

• feat(roles): per-role restart policy that relaunches an exited worker in place

  #451·ishandhanani opened Sep 15, 2026Collaborator·

  ·

• feat(dashboard): generate raw Tachometer UI for TRT-LLM runs

  #447·nv-yna opened Sep 14, 2026Collaborator·

  ·
  1

• fix(frontend): use Dynamo's two-tier worker-selection policy for KV routing

  #444·ishandhanani opened Sep 14, 2026Collaborator·

  ·
  1

• fix: release binaries from trusted main pushes

  #426·nv-yna opened Sep 12, 2026Collaborator·

  ·
  1

• [Power] support dedicated infrastructure with measured GPU power

  #424·edwingao28 opened Sep 11, 2026Contributor·

  ·

• [SA-Bench] preserve failed request results and measured boundaries

  #423·edwingao28 opened Sep 11, 2026Contributor·

  ·

• feat(profiling): nsight-slurm profiling type (Nsight Cloud for Slurm)

  #419·nv-yna opened Sep 11, 2026Collaborator·

  ·