All pull requests

Pull requests

• hip(RDNA4): a parity-gated fused-GR register-half down kernel for gfx1200/gfx1201 decode

  #2152·bsorensen110 opened Oct 12, 2026Contributor·

  ·
  4

• SYCL: preserve the borrowed default queue during teardown

  #2151·alex4o opened Oct 12, 2026·

  ·

• SYCL: opt in to 256 GRFs for fused hyperconnection up

  #2150·alex4o opened Oct 12, 2026·

  ·

• hip: add a gfx1150 hipBLASLt 100500 tuning table (Radeon 890M / Ryzen AI 9 HX 370)

  #2148·onionbutties opened Oct 12, 2026·

  ·
  1

• setup: opt-in gfx1150 (Radeon 890M / 880M, Strix Point) as a unified-memory AMD card

  #2147·onionbutties opened Oct 12, 2026·

  ·
  1

• sm_75: the v2 prompt attention in one stage, three blocks per SM (bitwise, attention 1.66x; stacked on #2092)

  #2146·rafatxf opened Oct 12, 2026Contributor·

  ·

• prefill: the fused int8 prompt experts on Turing (sm_75), opt-in (+9 % prefill on RTX 2080 Ti)

  #2145·rafatxf opened Oct 12, 2026Contributor·

  ·

• bench: community report, 2x Tesla P40 (PCIe 3.0 x8), i5-6500, 39 GiB RAM, IQ3_S, 131K, 0.1.43

  #2144·autemetry-james opened Oct 12, 2026·

  ·

• Community benchmark: mixed dual V100 + RTX PRO 4000 Blackwell on Windows

  #2142·intproduct opened Oct 12, 2026·

  ·

• serve: honour llama-server's thinking_budget_tokens (a llama.cpp client's thinking budget was silently ignored)

  #2140·xrip opened Oct 12, 2026·

  ·
  1

• Community benchmark: RTX 3060 on Windows, Strata 0.1.43, three quantizations through 260K input

  #2139·libratechw opened Oct 12, 2026Contributor·

  ·

• setup: reject conflicting --calibrate options (#2036)

  #2136·nekomario28 opened Oct 12, 2026Contributor·

  ·
  1

• bench: warn about expert profile contamination (#2088)

  #2135·nekomario28 opened Oct 12, 2026Contributor·

  ·
  1

• Community benchmark: Radeon AI PRO R9700 (gfx1201), ROCm 7.14.1, Ryzen 9 3900 (AVX2) — UD-IQ4_XS and the opt-in HIP switches, one at a time

  #2132·i-Vladimir-i opened Oct 12, 2026·

  ·

• serve: the prompt path's loan has room for a short request that keeps 0.1.39's ring (#2090)

  #2131·ischencheng opened Oct 12, 2026Contributor·

  ·
  1

• gitignore: local per-machine root files

  #2122·DassaultFalconKing opened Oct 12, 2026·

  ·

• Add disk-backed conversation cache parking (--conversation-cache-storage disk)

  #2119·dsovsyannikov opened Oct 12, 2026·

  ·
  1

• Community benchmark: RTX 5090 D v2 (24 GB), IQ3_S (abliterated), engine 0.1.43

  #2116·sobhechristl361 opened Oct 12, 2026·

  ·

• setup: the SYCL setup on an integrated Arc (xe): native launcher, a RAM fit rule, no KV streaming

  #2115·andersson82 opened Oct 12, 2026·

  ·

• sycl: run on an integrated Arc (Arc 140T, xe): free memory from /proc/meminfo, no prompt-slot borrowing, a native launcher

  #2114·andersson82 opened Oct 12, 2026·

  ·

• Community benchmark: RTX 4060 Ti 16 GB, Coder IQ1_M

  #2112·JoShMiQueL opened Oct 12, 2026·

  ·

• strata_bench: read CUDA UMD version, storage and GPU power/PCIe detail

  #2111·JoShMiQueL opened Oct 12, 2026·

  ·
  1

• sycl: STRATA_WINDOW_SYNC=1 by default on an integrated GPU (Arc 140T: 10 of 10 requests stalled, 20 of 20 pass)

  #2110·andersson82 opened Oct 12, 2026·

  ·

• --batch: request priority (a free slot and the next prompt read go to the highest priority waiting)

  #2109·brocktice opened Oct 12, 2026·

  ·

• sycl verify: complete STRATA_VERIFY_EAGER for the commit path

  #2106·demetree opened Oct 12, 2026Contributor·

  ·