Fusion

Workflow runs for Fusion

• cuda: raise CUDA graph cache cap from 64 to 512 for tensor parallel

  requires action with the application:
  Fusion#6:Pull request#232openedbyangliubc
  angliubc:fix/cuda-graph-cache-cap-tp

  Sep 23, 2026Action required
  More actions for workflow run 6

• MTP for Qwen3.8-Flash-Next

  cancelled:
  Fusion#5:Pull request#144synchronizebydanielhanchen
  mtp/qwen4exp-nextn

  Sep 18, 20264h 10m 25s
  More actions for workflow run 5

• qwen4exp: load the MTP head's hc_head_norm as [n_embd, hc] like the trunk gammas

  cancelled:
  Fusion#4:Pull request#219synchronizebydanielhanchen
  LeoBorcherding:fix/qwen4exp-nextn-hc-head-norm

  Sep 18, 20264h 14m 6s
  More actions for workflow run 4

• qwen4exp: load the MTP head's hc_head_norm as [n_embd, hc] like the trunk gammas

  cancelled:
  Fusion#3:Pull request#219synchronizebydanielhanchen
  LeoBorcherding:fix/qwen4exp-nextn-hc-head-norm

  Sep 18, 20265m 11s
  More actions for workflow run 3

• qwen4exp: load the MTP head's hc_head_norm as [n_embd, hc] like the trunk gammas

  cancelled:
  Fusion#2:Pull request#219openedbyLeoBorcherding
  LeoBorcherding:fix/qwen4exp-nextn-hc-head-norm

  Sep 17, 202614h 18m 36s
  More actions for workflow run 2

• MTP for Qwen3.8-Flash-Next

  cancelled:
  Fusion#1:Pull request#144synchronizebydanielhanchen
  mtp/qwen4exp-nextn

  Sep 16, 20264h 20m 26s
  More actions for workflow run 1