All pull requests

Pull requests

• [Engine] size the conversation cache for one snapshot of the configured context

  #4338·xvcezrqpbxtpzblrt opened Oct 27, 2026Contributor·

  ·

• [Engine] q36: a 4096-token prompt chunk in the hybrid mode when the card has room

  #4335·rynwang opened Oct 27, 2026·

  ·
  1

• [Engine] STRATA_HOST_TIMELINE host-timeline split + summary-time draft-profiler flush (opt-in, measurement only)

  #4334·xyzzing opened Oct 27, 2026Contributor·

  ·
  1

• [Engine] --mtp-q4 head/all fails at bind when no draft-head subset is installed

  #4333·xyzzing opened Oct 27, 2026Contributor·

  ·
  1

• Ship a CUDA 12 (sm_70) DeepSeek engine package so V100 users can download and run it

  needs-format
  #4330·unbug opened Oct 27, 2026·

  ·
  2

• [TUI] strata bench compare: say what a standard report is (Fix #4325)

  #4329·XiaoBinGan opened Oct 27, 2026·

  ·
  2

• [Engine] DeepSeek: send the medium effort to the template as 62

  #4327·nayuta-yanagisawa opened Oct 27, 2026·

  ·

• [Benchmark] tools/strata_bench.py --compare: say what a standard report is instead of KeyError (Fix #4325)

  #4326·XiaoBinGan opened Oct 27, 2026·

  ·

• [Benchmark] Community: one hour of back-to-back requests on a resident RTX 5090 (long soak)

  #4324·XiaoBinGan opened Oct 27, 2026·

  ·

• [Benchmark] community: one Intel Gaudi2 HL-225 (native backend in a fork), Qwen3.8-Flash-Next IQ4_XS, 262K

  #4322·JChen-318 opened Oct 27, 2026·

  ·
  1

• [Other] do not mark a whole shard as finished when its pinned hash is known (Fix #4088)

  #4318·XiaoBinGan opened Oct 27, 2026·

  ·
  1

• [TUI] a saved preset's lower-case model tag reaches setup as its key (Fix #4209)

  #4317·XiaoBinGan opened Oct 27, 2026·

  ·
  1

• [Docs] DOCKER.md: name the old-daemon start that looks like a card or image problem

  #4316·XiaoBinGan opened Oct 27, 2026·

  ·
  2

• [Engine] reject a boolean or a fractional token limit with 400 (Fix #4037)

  #4314·XiaoBinGan opened Oct 27, 2026·

  ·
  2

• [Benchmark] Community: RTX 5090 in a container on PCIe Gen 3 x16, Q2_0, with the card shared and with it exclusive

  #4313·XiaoBinGan opened Oct 27, 2026·

  ·
  3

• [TUI] launcher: start the bundled runtime when access(X_OK) is denied (Docker's default seccomp profile)

  #4312·XiaoBinGan opened Oct 27, 2026·

  ·
  2

• [GUI] Preserve image messages in provider chat history

  #4311·szl0834 opened Oct 27, 2026·

  ·

• [Engine] session files: format v2 with per-stage save/restore under --layer-split

  #4310·sakuumomo opened Oct 27, 2026Contributor·

  ·

• [Benchmark] RTX 2080 Ti 11 GB / Qwen3.8-Flash-Next / Gyro-S TQ1_0 + UD-IQ4_XS / strata-qwen38-thinking-v1-draft.1

  #4308·IlyaMolochnikov opened Oct 27, 2026·

  ·

• [Other] serve+webui: unload that works — pre-staged checkpoints, polite waits, honest answers

  #4307·sakuumomo opened Oct 27, 2026Contributor·

  ·
  2

• [Other] serve: session ladder — conversations survive an engine unload

  #4306·sakuumomo opened Oct 27, 2026Contributor·

  ·
  2

• [Engine] implement cache card logic for DeepSeek-V4.1, 6% faster decode

  #4302·Licenser opened Oct 27, 2026Contributor·

  ·

• [Engine] strata-dsv41: build with -Xptxas -dlcm=ca in the experimental sm_60/sm_70 build (Pascal decode 1.66 -> 3.50 tok/s)

  #4301·orlyandico opened Oct 27, 2026·

  ·
  1

• [Engine] strata-dsv41: build for Pascal (sm_61) by compiling the WMMA set kernels for sm_70+ only

  #4300·orlyandico opened Oct 27, 2026·

  ·
  1

• [Other] setup: keep the Unsloth env when a hipBLASLt table is found

  #4296·AMV007 opened Oct 27, 2026·

  ·
  2