All pull requests

Pull requests

• GGUF: open a metadata-only shard whose aligned data start falls past EOF (#1611)

  #1778·MegaRushX opened Oct 10, 2026·

  ·
  1

• serve: reasoning loop recovery also catches a loop written as one long word (#1753)

  #1777·BaaRai opened Oct 10, 2026·

  ·
  1

• sycl opt-in: STRATA_GR_FUSE_NORM=1, the hyper-connection read in 3 kernels instead of 4

  #1775·recutita opened Oct 10, 2026·

  ·

• sycl opt-in: STRATA_IQ3S_WIDE=1, a wide mmvq kernel for the dense IQ3_S matrices

  #1774·recutita opened Oct 10, 2026·

  ·

• bench: Q8 TCP parity and decode timings on 3x Tesla P4

  #1773·CC-David-CC opened Oct 10, 2026Contributor·

  ·

• prefill: Volta FP16 tensor-core prompt experts, with an IQ4_NL (type 20) down lane

  #1771·G0-0000 opened Oct 9, 2026·

  ·

• Windows AMD: image input on the CPU and on the GPU through Vulkan, te…

  #1768·WeisTekEng opened Oct 9, 2026·

  ·
  1

• sycl: add STRATA_DOORBELL_NOCHECK to skip the payload wait (debug only)

  #1767·duncanmcqueen opened Oct 9, 2026·

  ·

• sycl: add a device-event BF16/F32 MMVF bench

  #1766·duncanmcqueen opened Oct 9, 2026·

  ·

• docs: Intel Arc A770 decode speed investigation record

  #1765·duncanmcqueen opened Oct 9, 2026·

  ·

• EXL3: run the EXL3 trellis format on the engine

  #1764·Yxmura opened Oct 9, 2026·

  ·

• Add opt-in Galahad session storage (free non-commercial licence for 1 NVIDIA GPU)

  #1761·j-luwierski opened Oct 9, 2026Contributor·

  ·

• tools tests: test_amd_is_not_asked mocks the Linux HIP engine build too (it wrote sys.prefix/.strata-rocm.json)

  #1759·chaog992 opened Oct 9, 2026·

  ·

• bench: fleet calibration candidates and cache gains up to 22.6%

  #1755·CC-David-CC opened Oct 9, 2026Contributor·

  ·

• serve/web: add a favicon to the web app and the API Monitor

  #1754·Kurokosama opened Oct 9, 2026·

  ·

• HIP: unbuffered file-tier reads and a locked expert arena (#1691)

  #1752·KevinX8 opened Oct 9, 2026·

  ·

• bench: community report, RTX 5060 Ti 16 GB (WSL2), GSQ-RCO IQ3_S on engine 0.1.41

  #1751·kgmkm opened Oct 9, 2026·

  ·

• Two HIP fixes for Strix Halo (gfx1151): the wide top-k kernel and the VMM entry points

  #1749·vlupilin opened Oct 9, 2026·

  ·

• Remote stage: run the later layers of a layer split on other PCs, over TCP (opt-in; Linux and Windows; Qwen3.8-Flash-Next)

  #1748·lask3802 opened Oct 9, 2026·

  ·
  1

• bench: community report, 2x PH402 SKU 200 (4x GP100, sm_60), Swift 1.5 IQ3_XXS at 1M, --pipeline-windows off vs on over four stages, engine 0.1.41 + patches

  #1747·ruibeikaa opened Oct 9, 2026Contributor·

  ·

• SYCL: the PCIe probe copies from pinned host memory, as the CUDA engine's does (#1745)

  #1746·sam-fakhreddine opened Oct 9, 2026·

  ·
  1

• CUDA SM86: opt-in IQ3 two-stage prefill build

  #1744·InB4DevOps opened Oct 9, 2026·

  ·

• Gfx906 qsa fit17

  #1743·0FL01 opened Oct 9, 2026·

  ·

• CUDA SM86: opt-in one-superblock prefetch build

  #1742·InB4DevOps opened Oct 9, 2026·

  ·

• verify: service returns before any CUDA call while nothing has rung (#1740)

  #1741·architectds opened Oct 9, 2026Contributor·

  ·