All workflows

Workflow runs

• qwen CUDA: TF_DRAFT_NODES above 15 serves; a stop or cancel lands between prompt chunks

  completed successfully:
  No GPU#1249:Pull request#1078synchronizebyBobClawblaw
  BobClawblaw:q35-cuda-fixes

  Oct 22, 202611m 55s
  More actions for workflow run 1249

• qwen CUDA: TF_DRAFT_NODES above 15 serves; a stop or cancel lands between prompt chunks

  completed successfully:
  Zig format#1247:Pull request#1078synchronizebyBobClawblaw
  BobClawblaw:q35-cuda-fixes

  Oct 22, 202617s
  More actions for workflow run 1247

• qwen CUDA: TF_DRAFT_NODES above 15 serves; a stop or cancel lands between prompt chunks

  completed successfully:
  App and mobile checks#973:Pull request#1078synchronizebyBobClawblaw
  BobClawblaw:q35-cuda-fixes

  Oct 22, 20268m 33s
  More actions for workflow run 973

• glm/cuda: load and serve nvidia/GLM-5.3-Flash-FP4-Local as stored

  completed successfully:
  Zig format#1246:Pull request#1050synchronizebyBizuayeu
  Bizuayeu:pr/glm-fp4local

  Oct 22, 202618s
  More actions for workflow run 1246

• glm/cuda: load and serve nvidia/GLM-5.3-Flash-FP4-Local as stored

  completed successfully:
  No GPU#1248:Pull request#1050synchronizebyBizuayeu
  Bizuayeu:pr/glm-fp4local

  Oct 22, 202611m 40s
  More actions for workflow run 1248

• glm/cuda: load and serve nvidia/GLM-5.3-Flash-FP4-Local as stored

  completed successfully:
  App and mobile checks#972:Pull request#1050synchronizebyBizuayeu
  Bizuayeu:pr/glm-fp4local

  Oct 22, 20268m 57s
  More actions for workflow run 972

• qwen CUDA: --prompt-precision int8, prompt matmuls on int8 activations (cold prefill 20-27% shorter)

  completed successfully:
  Zig format#1245:Pull request#1176synchronizebyBobClawblaw
  BobClawblaw:q35-prompt-int8

  Oct 22, 202619s
  More actions for workflow run 1245

• qwen CUDA: --prompt-precision int8, prompt matmuls on int8 activations (cold prefill 20-27% shorter)

  completed successfully:
  App and mobile checks#971:Pull request#1176synchronizebyBobClawblaw
  BobClawblaw:q35-prompt-int8

  Oct 22, 20266m 47s
  More actions for workflow run 971

• qwen CUDA: --prompt-precision int8, prompt matmuls on int8 activations (cold prefill 20-27% shorter)

  completed successfully:
  No GPU#1247:Pull request#1176synchronizebyBobClawblaw
  BobClawblaw:q35-prompt-int8

  Oct 22, 202612m 38s
  More actions for workflow run 1247

• qwen CUDA: decode attention on tensor cores (fp8 and bf16 caches), decode after long prompts 1.5-2.3x faster

  completed successfully:
  Zig format#1244:Pull request#1168synchronizebyBobClawblaw
  BobClawblaw:q35-decode-attn-tc

  Oct 22, 202622s
  More actions for workflow run 1244

• qwen CUDA: decode attention on tensor cores (fp8 and bf16 caches), decode after long prompts 1.5-2.3x faster

  completed successfully:
  No GPU#1246:Pull request#1168synchronizebyBobClawblaw
  BobClawblaw:q35-decode-attn-tc

  Oct 22, 202616m 11s
  More actions for workflow run 1246

• qwen CUDA: decode attention on tensor cores (fp8 and bf16 caches), decode after long prompts 1.5-2.3x faster

  completed successfully:
  App and mobile checks#970:Pull request#1168synchronizebyBobClawblaw
  BobClawblaw:q35-decode-attn-tc

  Oct 22, 20269m 10s
  More actions for workflow run 970

• qwen CUDA: long-prompt prefill 5.4x faster (4096-row chunks, tensor-core attention, the delta-rule chain)

  completed successfully:
  No GPU#1245:Pull request#1080synchronizebyBobClawblaw
  BobClawblaw:q35-cuda-prefill

  Oct 22, 202614m 6s
  More actions for workflow run 1245

• qwen CUDA: long-prompt prefill 5.4x faster (4096-row chunks, tensor-core attention, the delta-rule chain)

  completed successfully:
  App and mobile checks#969:Pull request#1080synchronizebyBobClawblaw
  BobClawblaw:q35-cuda-prefill

  Oct 22, 20266m 56s
  More actions for workflow run 969

• qwen CUDA: long-prompt prefill 5.4x faster (4096-row chunks, tensor-core attention, the delta-rule chain)

  completed successfully:
  Zig format#1243:Pull request#1080synchronizebyBobClawblaw
  BobClawblaw:q35-cuda-prefill

  Oct 22, 202621s
  More actions for workflow run 1243

• qwen CUDA: Qwen3.8-27B decode 9-15% faster with DFlash2 (code +25%), replies unchanged

  completed successfully:
  Zig format#1242:Pull request#1079synchronizebyBobClawblaw
  BobClawblaw:q35-cuda-decode

  Oct 22, 202618s
  More actions for workflow run 1242

• qwen CUDA: Qwen3.8-27B decode 9-15% faster with DFlash2 (code +25%), replies unchanged

  completed successfully:
  App and mobile checks#968:Pull request#1079synchronizebyBobClawblaw
  BobClawblaw:q35-cuda-decode

  Oct 22, 20266m 36s
  More actions for workflow run 968

• qwen CUDA: Qwen3.8-27B decode 9-15% faster with DFlash2 (code +25%), replies unchanged

  completed successfully:
  No GPU#1244:Pull request#1079synchronizebyBobClawblaw
  BobClawblaw:q35-cuda-decode

  Oct 22, 202616m 16s
  More actions for workflow run 1244

• Maintainer fix-up for #1148: skip the table test off the pinned Unico…

  completed successfully:
  No GPU#1243:Commitbd26436pushedbyashhart
  main

  Oct 22, 202610m 23s
  More actions for workflow run 1243

• Maintainer fix-up for #1148: skip the table test off the pinned Unico…

  completed successfully:
  Zig format#1241:Commitbd26436pushedbyashhart
  main

  Oct 22, 202617s
  More actions for workflow run 1241

• Maintainer fix-up for #1148: skip the table test off the pinned Unico…

  completed successfully:
  App and mobile checks#967:Commitbd26436pushedbyashhart
  main

  Oct 22, 20268m 57s
  More actions for workflow run 967

• app: preserve model selections when reconnecting a provider

  requires action with the application:
  App and mobile checks#966:Pull request#1194openedbymonsterlag
  monsterlag:fix/issue-884-36d3

  Oct 22, 2026Action required
  More actions for workflow run 966

• app: preserve model selections when reconnecting a provider

  requires action with the application:
  Zig format#1240:Pull request#1194openedbymonsterlag
  monsterlag:fix/issue-884-36d3

  Oct 22, 2026Action required
  More actions for workflow run 1240

• app: preserve model selections when reconnecting a provider

  requires action with the application:
  No GPU#1242:Pull request#1194openedbymonsterlag
  monsterlag:fix/issue-884-36d3

  Oct 22, 2026Action required
  More actions for workflow run 1242

• glm/cuda: KV pages from the core pool, kept states sized by their position (#1142)

  completed successfully:
  Zig format#1239:Pull request#1163synchronizebyBizuayeu
  Bizuayeu:pr/glm-kv-pages

  Oct 22, 202618s
  More actions for workflow run 1239