All pull requests

Pull requests

• [Engine] Q4_K joins the expert down formats (Ornith-1.5-35B's 20 Q4_K down projections)

  #3679·yang478 opened Oct 22, 2026·

  ·

• [Engine] q36: quantized KV caches behind --kv (int8 / fp8 / rk4v4-e8 q4)

  #3678·yang478 opened Oct 22, 2026·

  ·

• [Benchmark] RTX 4070 Ti 12 GB + i7-13700K, 64 GB — Ornith-1.5-35B-A3B Q4_K_M on the q36 engine, --kv int8/fp8 at 150K/180K

  #3677·yang478 opened Oct 22, 2026·

  ·

• [Serve] a greedy request no longer inherits the model card's penalties (MTP stays on)

  needs-format
  #3676·yang478 opened Oct 22, 2026·

  ·
  1

• benchmark: 2x RX 7900 XTX / qwen3.8-flash-next / IQ3_S / strata-qwen38-thinking-v1-draft.1 (custom)

  needs-format
  #3672·matthewlenz opened Oct 22, 2026·

  ·
  1

• benchmark: 2x RX 7900 XTX / qwen3.8-flash-next / IQ3_S / strata-qwen38-thinking-v1-draft.1

  needs-format
  #3671·matthewlenz opened Oct 22, 2026·

  ·
  1

• [Docs] Share Companion Codex Reload Fix And Verified 200K Evidence

  #3668·medking82 opened Oct 22, 2026·

  ·
  3

• [Benchmark] Community Report: RX 7900 XTX / Qwen3.8-Flash-Next / IQ2_XS / 0.1.45

  #3665·blastpepsi1-eng opened Oct 22, 2026·

  ·

• [TUI] preserve log config and add indexed log rotation on launch

  #3663·TalenMud opened Oct 21, 2026·

  ·
  1

• [Engine] support Qwen3.8-Flash-Next Coder (256 experts) on Apple Silicon Metal

  #3662·chuaaron opened Oct 21, 2026·

  ·
  2

• [Benchmark] RTX 5080 / Qwen3.8-Flash-Next / UD-IQ4_XS / strata-qwen38-thinking-v1-draft.1

  #3658·Teracon00 opened Oct 21, 2026·

  ·
  3

• [Benchmark] RTX 5080 / Qwen3.8-Flash-Next / Q2_0 / strata-qwen38-thinking-v1-draft.1

  #3657·Teracon00 opened Oct 21, 2026·

  ·
  3

• [Benchmark] community report: Apple M1 Max, 32 GiB RAM, Q2_0 streaming

  #3650·demetree opened Oct 21, 2026Contributor·

  ·

• [Engine] --peer-device host route: STRATA_PF_PEER_PIECES=1 (opt-in) moves the activations and the sums in 8 pieces (+7 % prompt on 2x V100)

  #3648·fks opened Oct 21, 2026·

  ·

• [Benchmark] RX 6800 XT (gfx1030) / Flash-Next IQ3_S / ROCm 10.0 / v0.2.1 + v0.2.2 (vibe-coded)

  #3646·SpellsPT opened Oct 21, 2026·

  ·
  1

• [Other] root Dockerfile: pass the source commit to the terminal stage (source build fails at v0.2.2)

  #3644·robertoetcheverryr opened Oct 21, 2026·

  ·

• [Engine] Jev System One decision endpoint (/v1/systemone): typed decisions with calibrated probabilities

  #3643·justindressler opened Oct 21, 2026·

  ·

• bench: 2x RTX 2080 Ti, IQ3_S, 262K context, PCIe 3.0 x4

  needs-format
  #3642·geekjy opened Oct 21, 2026·

  ·
  1

• [Engine] opt-in qwen35moe F16 inference, validated on Sage-A1.5

  #3641·acrogenesis opened Oct 21, 2026Contributor·

  ·

• [Benchmark] Radeon AI PRO R9700 + Strix Halo (OCuLink, Linux) / Qwen3.8-Flash-Next / IQ3_S / strata-qwen38-thinking-v1-draft.1

  #3636·jiayeZhu opened Oct 21, 2026·

  ·

• [Benchmark] AMD Radeon AI PRO R9700 / Qwen3.8-Flash-Next IQ3_S / strata-qwen38-thinking-v1-draft.1

  #3633·l33tm4st3r opened Oct 21, 2026Contributor·

  ·
  3

• [Engine] STRATA_WINDOW_TRACE: a record per decode window (opt-in, record only)

  #3631·SanghunYun opened Oct 21, 2026Contributor·

  ·

• [Benchmark] bench.sh: name the AMD card on ROCm 7+, where only amd-smi exists

  #3630·l33tm4st3r opened Oct 21, 2026Contributor·

  ·
  4

• [Engine] fix the host memory probe on hybrid cgroup hosts and on WSL2 without systemd

  #3629·chaog992 opened Oct 21, 2026Contributor·

  ·

• [GUI] Add opt-in GPU energy and electricity cost monitoring to the Power card

  #3628·hash13 opened Oct 21, 2026Contributor·

  ·
  2