All workflows

Workflow runs

• Fused softmax_topk kernel

  requires action with the application:
  Build and Test#2885:Pull request#2054openedbypierreablin
  pierreablin:moe-routing-kernels

  Nov 12, 2026Action required
  More actions for workflow run 2885

• Fused Metal kernels for swiglu experts combination, faster single-token MoE experts

  requires action with the application:
  Build and Test#2884:Pull request#2053synchronizebypierreablin
  pierreablin:moe-fused-experts

  Nov 12, 2026Action required
  More actions for workflow run 2884

• Fused Metal kernels for swiglu experts combination, faster single-token MoE experts

  requires action with the application:
  Build and Test#2883:Pull request#2053openedbypierreablin
  pierreablin:moe-fused-experts

  Nov 12, 2026Action required
  More actions for workflow run 2883

• Server V2

  completed successfully:
  Build and Test#2882:Pull request#2040synchronizebymichalk8
  michalk8:serve-v2-pr

  Nov 11, 20263m 6s
  More actions for workflow run 2882

• Server V2

  completed successfully:
  Build and Test#2881:Pull request#2040synchronizebymichalk8
  michalk8:serve-v2-pr

  Nov 11, 20263m 52s
  More actions for workflow run 2881

• SwitchGLU: grouped-expert Metal kernels for small batches of 4-bit experts (1.4-1.7x at 32 concurrent requests)

  requires action with the application:
  Build and Test#2880:Pull request#2044synchronizebyTigerly1
  Tigerly1:moe-grouped-expert-kernels

  Nov 11, 2026Action required
  More actions for workflow run 2880

• Server V2

  completed successfully:
  Build and Test#2879:Pull request#2040synchronizebymichalk8
  michalk8:serve-v2-pr

  Nov 11, 20263m 47s
  More actions for workflow run 2879

• Server V2

  completed successfully:
  Build and Test#2878:Pull request#2040synchronizebymichalk8
  michalk8:serve-v2-pr

  Nov 11, 20263m 25s
  More actions for workflow run 2878

• Update bypass list

  completed successfully:
  Update bypass list#609:Scheduled
  main

  Nov 11, 202615s
  More actions for workflow run 609

• Add a harmony tool parser for gpt-oss in the server

  requires action with the application:
  Build and Test#2877:Pull request#2049openedbyVl4dk0
  Vl4dk0:harmony-tool-parser

  Nov 10, 2026Action required
  More actions for workflow run 2877

• Support tied word embeddings in GPT-OSS

  requires action with the application:
  Build and Test#2876:Pull request#2048openedbyAmitVaranasi
  AmitVaranasi:fix/gpt-oss-tie-word-embeddings

  Nov 10, 2026Action required
  More actions for workflow run 2876

• lora: say when the default model is used, and name the model being loaded

  requires action with the application:
  Build and Test#2875:Pull request#2047openedbySrikarmk
  Srikarmk:lora-name-the-model

  Nov 10, 2026Action required
  More actions for workflow run 2875

• Update bypass list

  completed successfully:
  Update bypass list#608:Scheduled
  main

  Nov 10, 202614s
  More actions for workflow run 608

• Fuse the DSA indexer's head sum

  requires action with the application:
  Build and Test#2874:Pull request#2046openedbyvinceclifford
  vinceclifford:indexer-head-sum

  Nov 9, 2026Action required
  More actions for workflow run 2874

• Update bypass list

  completed successfully:
  Update bypass list#607:Scheduled
  main

  Nov 9, 202611s
  More actions for workflow run 607

• server: apply configured adapter to model name

  requires action with the application:
  Build and Test#2873:Pull request#2033synchronizebyVikramansen
  Vikramansen:fix/adapter-path-model-alias

  Nov 9, 2026Action required
  More actions for workflow run 2873

• Update bypass list

  completed successfully:
  Update bypass list#606:Scheduled
  main

  Nov 8, 202617s
  More actions for workflow run 606

• perf(gated_delta): chunk-parallel differentiable training path for Ga…

  requires action with the application:
  Build and Test#2872:Pull request#2021synchronizebykru2710shna
  kru2710shna:fix/gated-delta-chunked-training

  Nov 8, 2026Action required
  More actions for workflow run 2872

• SwitchGLU: grouped-expert Metal kernels for small batches of 4-bit experts (1.4-1.7x at 32 concurrent requests)

  requires action with the application:
  Build and Test#2871:Pull request#2044openedbyTigerly1
  Tigerly1:moe-grouped-expert-kernels

  Nov 7, 2026Action required
  More actions for workflow run 2871

• server: reject empty prompts instead of ending the generation thread

  completed successfully:
  Update bypass list#605:Pull request#2043closedbyTigerly1
  Nov 7, 202616s
  More actions for workflow run 605

• server: reject empty prompts instead of ending the generation thread

  failed:
  Build and Test#2870:Pull request#2043openedbyTigerly1
  Tigerly1:server-reject-empty-prompt

  Nov 7, 202642m 43s
  More actions for workflow run 2870

• fix(gpt_oss): detect transformers [E, H, 2F] layout in sanitize() (fixes #1994)

  requires action with the application:
  Build and Test#2869:Pull request#2009synchronizebykru2710shna
  kru2710shna:fix/gpt-oss-sanitize-layout-c

  Nov 7, 2026Action required
  More actions for workflow run 2869

• perf(gated_delta): chunk-parallel differentiable training path for Ga…

  requires action with the application:
  Build and Test#2868:Pull request#2021synchronizebykru2710shna
  kru2710shna:fix/gated-delta-chunked-training

  Nov 7, 2026Action required
  More actions for workflow run 2868

• Add NVIDIA Nemotron-Labs Dory model

  requires action with the application:
  Build and Test#2867:Pull request#1988synchronizebypmichalas
  pmichalas:add-dory

  Nov 7, 2026Action required
  More actions for workflow run 2867

• Fix RotatingKVCache mask hiding keep (sink) slots during chunked prefill

  requires action with the application:
  Build and Test#2866:Pull request#2041synchronizebyrafiimam
  rafiimam:fix/rotating-cache-sink-mask

  Nov 7, 2026Action required
  More actions for workflow run 2866