All pull requests

Pull requests

• [Benchmark] AMD Radeon RX 7900 XT - qwen3.8-flash-next iq3_s

  #4061·fsacrini opened Oct 25, 2026·

  ·
  2

• [Other] serve: logprobs on /v1/completions (stacks #3200 + #3332)

  #4056·gdevenyi opened Oct 25, 2026·

  ·

• Community benchmark: RTX 3090 + i9-10900K, Flash-Next IQ3_S at 512K context (engine 0.1.48) + seven-configuration A/B on engine 0.1.44.1

  needs-format
  #4054·smallmj opened Oct 25, 2026·

  ·
  1

• [Engine] alias_sampling: request defaults per model alias

  #4050·furkangkhsn opened Oct 25, 2026·

  ·

• [Engine] replicas: each replica its own server on its own port, with one shared expert arena

  #4049·furkangkhsn opened Oct 25, 2026·

  ·

• [Engine] add opt-in reference-order CPU mixed-IQ down dispatch

  #4046·CrashxZ opened Oct 25, 2026·

  ·
  1

• [Engine] --lookup-max-t: prompt-lookup windows up to 16 tokens while the MTP keeps --spec, bitwise, edits +19% decode

  #4040·merbanan opened Oct 25, 2026Contributor·

  ·

• [Engine] opt-in smaller prompt scratch (STRATA_PF_GROUP_SCRATCH, STRATA_PF_HC_REGION): bigger --prefill auto chunks, bitwise, 8 GB cards +18-24% prompt

  #4039·merbanan opened Oct 25, 2026Contributor·

  ·

• [Engine] Conversation cache: Stop evicting what this freed RAM can never re-use; Auto-shrinking RAM conversation cache slots under RAM pressure

  #4031·dsenchankau opened Oct 25, 2026·

  ·

• [Other] SYCL CMakeLists.txt: Port missing compiler flags from commit 1747c8e53

  #4030·sruggier opened Oct 25, 2026·

  ·

• [Engine] apply the existing sm_120 default and fix the native build on RTX 50

  #4029·pschaub opened Oct 25, 2026·

  ·
  1

• [Docs] SAFETY.md: say that strata launch updates the DeepSeek engine itself

  #4028·pschaub opened Oct 25, 2026·

  ·
  1

• [Other] setup: let the server validate component-backed engines

  #4027·jubeater opened Oct 25, 2026·

  ·
  1

• [Engine] generate: allow the elastic K/V with --peer-device

  #4020·chimpera opened Oct 24, 2026·

  ·

• [Engine] Mac: calibrate the decode kernels' tiles on the device (setup.sh --calibrate)

  #4019·rasa04 opened Oct 24, 2026·

  ·

• [GUI] Update and restart from the server's web page (strata update, then the same server again)

  #4018·elightcap opened Oct 24, 2026·

  ·

• [Engine] dsv41: --numa-experts bands — every NUMA node serves the CPU lane

  #4016·chimpera opened Oct 24, 2026·

  ·

• [Benchmark] standard draft pilot: RX 6900 XT 16 GB on Windows 11 / Qwen3.8-Flash-Next / IQ3_S / strata-qwen38-thinking-v1-draft.1 (9 runs)

  #4012·Vop3p opened Oct 24, 2026Contributor·

  ·

• [Benchmark] tools/strata_bench.py: read CMakeLists.txt as UTF-8, not the locale's code page (fixes #2792)

  #4011·SuperDistance opened Oct 24, 2026Contributor·

  ·
  1

• [Engine] Pipelined batch groups adapt the VRAM tier (opt-in STRATA_PIPE_ADAPT_EVERY): six-stream decode after a start 309 -> 410 tok/s

  #4010·SuperDistance opened Oct 24, 2026Contributor·

  ·

• [Engine] prefill timing: do not record a mark in the phase already running (the breakdown charged copy waits to "dequant" on Windows HIP)

  #4006·Vop3p opened Oct 24, 2026Contributor·

  ·
  1

• [Engine] prefill: opt-in sparse copied events and deduplicated slot waits (Windows HIP prompt reads +36% at 4K, +15% at 32K)

  #4005·Vop3p opened Oct 24, 2026Contributor·

  ·
  1

• [Benchmark] RTX 5090 / Qwen3.8-Flash-Next NVFP4 / strata-qwen38-thinking-v1-draft.1

  #4003·HOOD233233 opened Oct 24, 2026·

  ·

• [Engine] Parallelize staged conversation cache copies

  #4002·oscarlius opened Oct 24, 2026·

  ·

• [Engine] build the DeepSeek engine for sm_89 (the RTX 40 series)

  #3994·woodsonl opened Oct 24, 2026·

  ·