Release

Workflow runs for Release

• cuda : fuse qwen4exp's SCALE -> SIGMOID -> SCALE -> hc_post chain (#3…

  waiting for another serialized run to finish:
  Release#289:Commitff5a233pushedbyEAddario
  master

  Oct 18, 2026Pending
  More actions for workflow run 289

• hexagon: vectorize softplus with HVX (#30435)

  waiting for another serialized run to finish:
  Release#288:Commit27c7a76pushedbyEAddario
  master

  Oct 16, 2026Pending
  More actions for workflow run 288

• opencl: Add MxFP4 MoE bin kernel for Adreno X1 devices (#30691)

  waiting for another serialized run to finish:
  Release#287:Commit2f2763dpushedbyEAddario
  master

  Oct 16, 2026Pending
  More actions for workflow run 287

• model : Granite Switch Decision (#30676)

  waiting for another serialized run to finish:
  Release#286:Commit23f7aabpushedbyEAddario
  master

  Oct 16, 2026Pending
  More actions for workflow run 286

• common: honor HTTP_PROXY/HTTPS_PROXY env vars in http client (#21752)

  queued:
  Release#285:Commit7182ef7pushedbyEAddario
  master

  Oct 16, 2026Queued
  More actions for workflow run 285

• model : MTP support for Mellum2.1 (#30612)

  cancelled:
  Release#284:Commit9a16f1dpushedbyEAddario
  master

  Oct 16, 20262d 5h 6m 47s
  More actions for workflow run 284

• vulkan: Optimize pool_2d when the pooling window is large (#30119)

  cancelled:
  Release#283:Commit77c4540pushedbyEAddario
  master

  Oct 13, 20263d 21h 57m 25s
  More actions for workflow run 283

• cuda : fix compiler warnings in mmq-load-tiles.cuh (#30443)

  failed:
  Release#282:Commitbfe65e1pushedbyEAddario
  master

  Oct 13, 20263d 1h 22m 22s
  More actions for workflow run 282

• convert : fix LoRA conversion with --outtype auto (#30433)

  cancelled:
  Release#281:Commit9d7d504pushedbyEAddario
  master

  Oct 13, 20263d 4h 24m 6s
  More actions for workflow run 281

• cuda: always use MMVQ for MUL_MAT_ID on sm_60 (#27828)

  cancelled:
  Release#280:Commit1623d8cpushedbyEAddario
  master

  Oct 10, 20265d 6h 11m 5s
  More actions for workflow run 280

• vendor: apply deep nested json patch from upstream (#30253)

  cancelled:
  Release#279:Commit10a60cfpushedbyEAddario
  master

  Oct 9, 20264d 15h 1m 10s
  More actions for workflow run 279

• CUDA: fix round issue, under MSVC the CPU and GPU agree (#30229)

  cancelled:
  Release#278:Commit79e2e74pushedbyEAddario
  master

  Oct 9, 20263d 15h 38m 49s
  More actions for workflow run 278

• CUDA: pass src1 precision to host MMQ config helpers (#30168)

  cancelled:
  Release#277:Commit8a1a9b5pushedbyEAddario
  master

  Oct 9, 20263d 6h 16m 45s
  More actions for workflow run 277

• sycl: FWHT optimizations (#29605)

  cancelled:
  Release#276:Commit46baf1fpushedbyEAddario
  master

  Oct 8, 20263d 5h 58m 31s
  More actions for workflow run 276

• chat: fix jinja parser for TranslateGemma (#30096)

  failed:
  Release#275:Commitbd4eeaapushedbyEAddario
  master

  Oct 7, 20262d 16h 7m 15s
  More actions for workflow run 275

• Release

  failed:
  Release#274:byEAddario
  master

  Oct 7, 20262d 20h 2m 48s
  More actions for workflow run 274

• feat: add GLM5Next MTP, optimize (#29928)

  failed:
  Release#273:Commitb9acf13pushedbyEAddario
  master

  Oct 7, 20263d 1h 17m 48s
  More actions for workflow run 273

• ggml-cuda: use per-thread stream for buffer-init padding memset (#28782)

  cancelled:
  Release#272:Commit51ce9c1pushedbyEAddario
  master

  Oct 6, 20263d 17h 21m 57s
  More actions for workflow run 272

• ci : winget urls must be separate strings (#29978)

  cancelled:
  Release#271:Commite5983d6pushedbyEAddario
  master

  Oct 5, 20264d 4h 54m 40s
  More actions for workflow run 271

• ggml-cpu: support BF16/FP16/FP32 K tails in tinyBLAS on x86 (#29806)

  failed:
  Release#270:Commita7b94dfpushedbyEAddario
  master

  Oct 4, 20263d 16h 13m 14s
  More actions for workflow run 270

• CUDA: fix MMQ memory fault if n_expert >> n_ubatch (#29941)

  failed:
  Release#269:Commitdd26678pushedbyEAddario
  master

  Oct 4, 20264d 0h 9m 45s
  More actions for workflow run 269

• imatrix: calculate activation-based statistics for new format (GGUF) …

  failed:
  Release#268:Commit0504396pushedbyEAddario
  master

  Oct 4, 20264d 1h 14m 29s
  More actions for workflow run 268

• model : Add LFM2.5-Encoder-350M and LFM2.5-Encoder-230M (#29862)

  cancelled:
  Release#267:Commitcb7934cpushedbyEAddario
  master

  Oct 3, 20265d 5h 26m 8s
  More actions for workflow run 267

• CUDA: fuse shared experts into MMVQ (#29184)

  cancelled:
  Release#266:Commitbed0a85pushedbyEAddario
  master

  Oct 2, 20264d 15h 11m 45s
  More actions for workflow run 266

• metal : add tensor API flash attention kernel for F16 KV (#29570)

  cancelled:
  Release#265:Commit926862epushedbyEAddario
  master

  Oct 2, 20264d 1h 17m 14s
  More actions for workflow run 265