Behavior Check

Workflow runs for Behavior Check

• pr-set: update

  cancelled:
  Behavior Check#17:Pull request#336synchronizebydanielhanchen
  pr-set-update

  Oct 29, 20265m 13s
  More actions for workflow run 17

• pr-set: update

  cancelled:
  Behavior Check#16:Pull request#336openedbydanielhanchen
  pr-set-update

  Oct 29, 20268m 24s
  More actions for workflow run 16

• llama : flag the outputs of every backend sampling chain

  completed successfully:
  Behavior Check#15:Pull request#335openedbydanielhanchen
  fix/backend-sampling-outputs

  Oct 29, 20269m 37s
  More actions for workflow run 15

• model : DeepSeek-V4.1-Flash-1017 pin on upstream deepseek41 (MXFP8 Engram + extras)

  completed successfully:
  Behavior Check#14:Pull request#332openedbydanielhanchen
  dsv41-mxfp8-pin

  Oct 27, 202616m 4s
  More actions for workflow run 14

• metal: add NVFP4 mul_mat, mul_mat_id and get_rows kernels

  requires action with the application:
  Behavior Check#13:Pull request#330openedbyLyxot
  Lyxot:fix/metal-nvfp4

  Oct 27, 2026Action required
  More actions for workflow run 13

• vulkan: copy a misaligned top_k input to an aligned scratch instead of aborting

  completed successfully:
  Behavior Check#12:Pull request#328synchronizebydanielhanchen
  fix/vulkan-topk-misaligned-src

  Oct 26, 20263m 35s
  More actions for workflow run 12

• vulkan: copy a misaligned top_k input to an aligned scratch instead of aborting

  completed successfully:
  Behavior Check#11:Pull request#328synchronizebydanielhanchen
  fix/vulkan-topk-misaligned-src

  Oct 26, 20261m 48s
  More actions for workflow run 11

• IQ1_XS, IQ1_XXS, IQ1_XXXS: three quant types below IQ1_S

  completed successfully:
  Behavior Check#10:Pull request#61reopenedbydanielhanchen
  iq1-narrow

  Oct 25, 202610m 47s
  More actions for workflow run 10

• CUDA: add IQ1_M to MMQ

  completed successfully:
  Behavior Check#9:Pull request#304synchronizebydanielhanchen
  cuda-iq1m-mmq

  Oct 25, 202610m 38s
  More actions for workflow run 9

• CUDA: faster MoE prompt processing (compact MMQ tiling, device expert sort, GET_ROWS)

  completed successfully:
  Behavior Check#8:Pull request#264synchronizebydanielhanchen
  cuda-moe-prompt-kernels

  Oct 25, 20264m 26s
  More actions for workflow run 8

• model : DeepSeek-V4.1-Flash-1017 full stack pin (all DeepSeek-V4.1 PRs except #292)

  failed:
  Behavior Check#7:Pull request#298synchronizebydanielhanchen
  dsv41-all

  Oct 25, 20266m 0s
  More actions for workflow run 7

• IQ1_XS, IQ1_XXS, IQ1_XXXS: three quant types below IQ1_S

  completed successfully:
  Behavior Check#6:Pull request#61synchronizebydanielhanchen
  iq1-narrow

  Oct 25, 20269m 15s
  More actions for workflow run 6

• model: add MiniMax-M3.1

  completed successfully:
  Behavior Check#5:Pull request#320synchronizebydanielhanchen
  minimax-m31

  Oct 25, 20264m 12s
  More actions for workflow run 5

• model : DeepSeek-V4.1-Flash-1017 full stack pin (all DeepSeek-V4.1 PRs except #292)

  failed:
  Behavior Check#4:Pull request#298synchronizebydanielhanchen
  dsv41-all

  Oct 25, 20267m 22s
  More actions for workflow run 4

• Carry what the dropped GLM-5-Next, Qwen MTP and readahead pins had beyond upstream

  completed successfully:
  Behavior Check#3:Pull request#241synchronizebydanielhanchen
  carry/superseded-pin-optimizations

  Oct 24, 20269m 58s
  More actions for workflow run 3

• llama-mmap: populate the non-lazy ranges when MAP_POPULATE is skipped for lazy tensors

  completed successfully:
  Behavior Check#2:Pull request#266synchronizebydanielhanchen
  mmap-populate-nonlazy

  Oct 23, 202633m 3s
  More actions for workflow run 2

• server: preemption notices, asynchronous parks and exact concurrency (pin branch on b11577)

  completed successfully:
  Behavior Check#1:Pull request#269synchronizebydanielhanchen
  pin/server-side-preemption-b11577

  Oct 23, 202625m 33s
  More actions for workflow run 1