All pull requests

Pull requests

• fix: align SGLang admission and LRU boundary behavior

  #571·xuhuan51 opened Oct 23, 2026·

  ·
  17

• fix: adapt vLLM attention metadata to removed CPU fields

  #569·simone-chen opened Oct 22, 2026Contributor·

  ·
  1

• fix: qualify TRT-LLM rc29 collectors for Dynamo 1.6

  #568·simone-chen opened Oct 22, 2026Contributor·

  ·
  1

• fix: align learned FPM defaults with gym recipes

  #567·diffractiongoo opened Oct 22, 2026Contributor·

  ·
  8

• fix: defer unknown cache layouts to the runtime probe in onboarding

  #565·Arsene12358 opened Oct 22, 2026Contributor·

  ·
  3

• fix: release agentic profile requests from dependency events like AIPerf

  #564·dreamtalen opened Oct 22, 2026Contributor·

  ·
  9

• fix: refresh bundled zai-org GLM-5.3 configs to the published checkpoints

  #563·Arsene12358 opened Oct 22, 2026Contributor·

  ·
  1

• feat: accept native CUDA timings for autoregressive execution

  #561·milesial opened Oct 22, 2026Contributor·

  ·
  1

• feat: support session-pinned aggregated and disaggregated replay

  #560·milesial opened Oct 22, 2026Contributor·

  ·
  6

• feat: constrain recommendations by request latency p99

  #556·liangqianxing opened Oct 22, 2026·

  ·
  2

• fix: include context parallelism in the agg worker setup table

  #549·v0ropaev opened Oct 21, 2026·

  ·
  3

• fix: check saved CP rows against the generated config version

  #548·v0ropaev opened Oct 21, 2026·

  ·
  2

• ci: check bundled model config provenance

  #547·saturley-hall opened Oct 21, 2026Member·

  ·
  6

• fix: reject unsupported Task serving modes at construction

  #546·Arsene12358 opened Oct 21, 2026Contributor·

  ·
  3

• fix: reject unsupported pipeline parallelism in engine replay

  #545·milesial opened Oct 21, 2026Contributor·

  ·
  1

• fix: accept native dense KV skip evidence in FPM collection

  #529·simone-chen opened Oct 20, 2026Contributor·

  ·
  2

• fix: preserve load-match ranking across backends

  #524·git-jxj opened Oct 20, 2026·

  ·
  21

• feat: support vLLM G2 host offload with MTP

  #519·Harrilee opened Oct 19, 2026Contributor·

  ·
  7

• data: 2026-10 campaign — vLLM 0.30.0 / TRT-LLM 1.3.0rc29 / SGLang 0.5.21 perf tables for 8 systems

  #517·tianhaox opened Oct 19, 2026Contributor·

  ·
  5

• feat(afd): measure the AFD MoE stage from FastAFD profiles

  #510·liz-badada opened Oct 19, 2026·

  ·
  6

• fix: charge the prefill logits GEMM for MoE-family models

  #509·PeaBrane opened Oct 18, 2026Contributor·

  ·
  1

  5

• fix: correct decode QK-norm latency accounting

  #506·jontb opened Oct 17, 2026·

  ·
  1

  9

• fix: accept canonical speculation in engine launch adapters

  #505·Harrilee opened Oct 17, 2026Contributor·

  ·
  3

• fix: explain pure-TP MoE KV warm-up runtime requirements

  #501·simone-chen opened Oct 16, 2026Contributor·

  ·
  5

• test: isolate real torch imports across collector tests

  #497·git-jxj opened Oct 16, 2026·

  ·
  3