All pull requests

Pull requests

• [GUI] DeepSeek load/tune keeps --vision on|off and drops only a bare --vision

  #4572·Nyarlathoteppppp opened Oct 29, 2026·

  ·
  1

• [Other] update.sh / UPDATE.bat: a failed pull without origin/main is not the old history

  #4571·Nyarlathoteppppp opened Oct 29, 2026·

  ·
  1

• [GUI] Time Skip Stop ends the job's process group on Linux and macOS

  #4570·Nyarlathoteppppp opened Oct 29, 2026·

  ·
  1

• [GUI] Image-model downloads honour HF_ENDPOINT like setup and the MTP fetch

  #4569·Nyarlathoteppppp opened Oct 29, 2026·

  ·
  1

• [TUI] strata launch --stop: a failed stop no longer says nothing is running, and exits 1

  #4568·Nyarlathoteppppp opened Oct 29, 2026·

  ·
  1

• [Engine] QSA decode attention: opt-in 32-cell chunks (STRATA_QSA_ATTN_VARIANT=2)

  needs-format
  #4567·Jackwwg83 opened Oct 29, 2026·

  ·
  1

• bench(results): Jetson AGX Thor (sm_110, aarch64) IQ3_S standard run

  needs-format
  #4565·potoior opened Oct 29, 2026·

  ·
  1

• [TUI] DeepSeek start_env test: assert on the library variable the code writes

  #4564·Nyarlathoteppppp opened Oct 29, 2026·

  ·
  1

• [TUI] kept_key: skip unreadable profiles instead of minting a new API key

  #4563·Nyarlathoteppppp opened Oct 29, 2026·

  ·
  1

• [TUI] running_launches: an unreadable launch profile no longer reports the previous launch's config

  #4562·Nyarlathoteppppp opened Oct 29, 2026·

  ·
  1

• [Benchmark] RTX 4090 + RTX 5060 Ti, UD-IQ4_XS on DDR5-4800

  #4558·ayvan opened Oct 29, 2026·

  ·

• [Benchmark] RTX 5080 + RTX 5060 Ti dual-GPU: four draft-standard Flash-Next runs and 15 legacy results

  #4555·bubbleofmoon opened Oct 29, 2026·

  ·

• [Engine] session-file v3: incremental append saves under --session-save-append

  #4551·sakuumomo opened Oct 28, 2026Contributor·

  ·
  2

• [Engine] layer split: keep the draft head's room out of the last card's cache (from Strata-DualGPU, #642)

  #4549·cobra91 opened Oct 28, 2026·

  ·
  1

• [Benchmark]: RX 7600 8 GB HIP Qwen3.8-Flash-Next Q2_0 / Qwen3.8-Flash-Next Coder IQ1_M

  needs-format
  #4543·ominiverdi opened Oct 28, 2026·

  ·
  1

• [Benchmark] 2x RTX 4060 Ti 16 GB / Qwen3.8-Flash-Next UD-IQ4_XS / v0.2.7.1: production config and the #3888 flags

  #4539·fioruccione opened Oct 28, 2026Contributor·

  ·

• [Engine] preserve periodic retry after a low-benefit prefetch closure

  #4538·cmdruid opened Oct 28, 2026·

  ·
  1

• [Engine] gguf_reader: header bounds check cannot wrap on a huge string length

  #4536·schmittcn opened Oct 28, 2026·

  ·

• [GUI] Shared listener: compare the API key in constant time (as the engine since #213)

  #4535·schmittcn opened Oct 28, 2026·

  ·

• [Engine] split search: price a pipelined two-stage split by its slower stage (from Strata-DualGPU, #642)

  needs-format
  #4534·cobra91 opened Oct 28, 2026·

  ·
  2

• [Benchmark] RTX 3080 Ti + RTX 3080 / Qwen3.8 Flash Next IQ3_S / draft.1

  #4527·maxlippe-leonardo opened Oct 28, 2026·

  ·

• [Engine] layer split: trim every card's dense weights on --layer-split auto too (from Strata-DualGPU, #642)

  #4522·cobra91 opened Oct 28, 2026·

  ·
  5

• [Benchmark] 2x RTX 2080 Ti on a slow dual-socket Xeon, engine 0.1.52: split vs one card, and IQ3_XXS

  #4515·NorwinYu opened Oct 28, 2026Contributor·

  ·

• [Benchmark] run.command is shell-quoted, so a reader can paste it back (#4494)

  needs-format
  #4513·ctr-137 opened Oct 28, 2026Contributor·

  ·
  1

• [Benchmark] Add community benchmark reports: 2x RTX 3060, Qwen3.8-Flash-Next IQ3_S and Unsloth UD-IQ4_XS (2026-10-28)

  #4508·iskrid opened Oct 28, 2026·

  ·
  2