All pull requests

Pull requests

• feat: honor exporter dependencies in Dynamo request traces

  #380·PeaBrane opened Oct 3, 2026Contributor·

  ·
  10

• fix: harden FPM collector completion and failure handling

  #379·Arsene12358 opened Oct 3, 2026Contributor·

  ·
  2

• fix(replay): withhold synthesized trace session IDs from placement

  #378·PeaBrane opened Oct 3, 2026Contributor·

  ·
  1

• feat: migrate FPM Gym dashboards to github Pages

  #375·simone-chen opened Oct 2, 2026Contributor·

  ·
  9

• fix: align InferenceX source resolution and E2E accuracy with gym

  #372·simone-chen opened Oct 2, 2026Contributor·

  ·
  4

• DeepSeek-V4.1 operator profiles on H20: sglang 0.5.21 + vLLM 0.30.0 producers (on #228)

  #369·tianhaox opened Oct 2, 2026Contributor·

  ·
  2

• feat: add op-based E2E overview and details views

  #368·simone-chen opened Oct 2, 2026Contributor·

  ·
  7

• Fix: Improving KV Cache token count accuracy in Sliding Window Attention and NVFP4 models for vLLM

  #366·mazam-lab opened Oct 1, 2026·

  ·
  6

• refactor: use SGLang native radix core for simulation

  #363·ishandhanani opened Oct 1, 2026·

  ·
  3

• ci: fix nightly license baseline downloads

  #362·saturley-hall opened Oct 1, 2026Member·

  ·
  4

• feat: add GLM-5.3-Flash standalone per-op collection and SILICON tables on GB300

  #361·Harrilee opened Oct 1, 2026Contributor·

  ·
  1

• docs: link Hugging Face FPM dataset and contributor guide

  #357·simone-chen opened Oct 1, 2026Contributor·

  ·
  4

• ci: publish license evidence for PR and post-merge runs

  #355·saturley-hall opened Sep 30, 2026Member·

  ·
  1

• build: remove transitive Thrift through Parquet upgrade

  #350·saturley-hall opened Sep 30, 2026Member·

  ·
  1

• fix: pin DSV4.1 FPM data to schema-valid HF revision

  #345·Harrilee opened Sep 29, 2026Contributor·

  ·
  2

• fix: stop inflating Weka trace memory estimates

  #341·abhibambhaniya opened Sep 29, 2026·

  ·
  10

• fix: count only uncached tokens against the legacy agg prefill budget

  #333·kangclzjc opened Sep 28, 2026·

  ·
  20

• feat(sglang): model PD decode admission headroom

  #332·daii-0818 opened Sep 28, 2026·

  ·
  8

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

  #331·PeaBrane opened Sep 27, 2026Contributor·

  ·
  1

• feat: enable B60 ops collection in xpu-graph mode and XPU-calibrated vLLM agg backend

  #329·mandy-li opened Sep 26, 2026·

  ·
  58

• feat: add GLM-5.3-Flash FPM collection and GB300 profiles

  #324·Harrilee opened Sep 23, 2026Contributor·

  ·
  1

• feat: add GLM-5.3-Flash text architecture and SOL support

  #323·Harrilee opened Sep 23, 2026Contributor·

  ·
  1

• refactor: drop legacy AFD interfaces

  #320·LPhgh opened Sep 23, 2026·

  ·
  26

• ci: add advisory simulation performance gate

  #318·jthomson04 opened Sep 23, 2026Contributor·

  ·
  43

• feat: support worker context limits in recommendations

  #312·natoscott opened Sep 22, 2026·

  ·
  10