All pull requests

Pull requests

• [GUI] keep the API model when New chat opens a chat

  #5750·hexgrimm opened Nov 10, 2026·

  ·

• [Engine] verify: STRATA_TAP_FILE works on a layer split

  #5748·botBehavior opened Nov 10, 2026·

  ·

• [Benchmark] RTX PRO 4000 Blackwell / Qwen3.8-Flash-Next / NVFP4 GPTQ / strata-qwen38-thinking-v1-draft.1: 64 GB (simulated) vs 128 GB RAM

  #5746·Adam90a opened Nov 10, 2026·

  ·
  2

• [Engine] serve drafting: Intel starts lose DFlash2 because the check reads strata-sycl.sh, not its binary (#5138)

  #5745·Fett2 opened Nov 10, 2026·

  ·

• [Engine] let gfx906 build and run the DeepSeek V4.1 engine

  #5739·shjszwh opened Nov 10, 2026·

  ·
  3

• [Benchmark] 0.2.13.1 DeepSeek-V4.1-Flash 2.00bpw on an RTX 5060 Ti 16 GB / 64 GB (Atlas)

  #5736·jkenner111 opened Nov 10, 2026·

  ·

• [Benchmark] Qwen3.8-Flash-Next IQ2_XS on one RTX 5080

  #5735·Hotziplotzi opened Nov 10, 2026·

  ·

• [Benchmark] RTX 5080 / Qwen3.8-Flash-Next / UD-Q4_K_XL on 0.2.13.1

  #5733·Hotziplotzi opened Nov 10, 2026·

  ·

• [Benchmark] RTX 5080 / Qwen3.8-Flash-Next / IQ3_XXS on 0.2.13.1

  #5732·Hotziplotzi opened Nov 10, 2026·

  ·

• [Benchmark] 0.2.13.1 DeepSeek-V4.1-Flash 2.00bpw on two mismatched Blackwell cards (HP Z820, AVX1)

  #5731·jkenner111 opened Nov 10, 2026·

  ·

• [Benchmark] RTX 5080 / Qwen3.8-Flash-Next / IQ3_S on 0.2.13.1

  #5728·Hotziplotzi opened Nov 10, 2026·

  ·

• [Engine] enable multi-step MTP across concurrent fixed slots

  #5726·shoucandanghehe opened Nov 10, 2026Contributor·

  ·

• [Engine] gfx1152: enable the matrix-core prompt path (runtime gates + WMMA device guards)

  #5725·chadstthomas-svg opened Nov 10, 2026·

  ·
  1

  2

• [GUI] Fix MCP tool forwarding in Desktop Serving API

  #5721·vladare opened Nov 10, 2026·

  ·
  1

• [Engine] fix dsv41 on Pascal

  #5713·rafal-prasal opened Nov 10, 2026Contributor·

  ·
  2

• [Engine] Add opt-in NUMA-aware idle CPU expert prefetch

  #5712·brokenarrow2099 opened Nov 10, 2026·

  ·

• [Benchmark] Community bench: RTX 4080 SUPER / Ryzen 7950X3D - Flash-Next IQ3_XXS, IQ3_S and Swift on engine 0.1.59

  #5709·c-OOL-io opened Nov 10, 2026·

  ·

• [Engine] fix the --draft-conf slot estimate for UD-Q4_K_XL and other native-experts packs

  #5700·woodsonl opened Nov 10, 2026·

  ·

• [Engine] session-identity: the whole STRATA_* environment and the device plan join the config fingerprint (#5379)

  #5699·nazerim opened Nov 10, 2026·

  ·
  2

• [Engine] opt-in vision.elastic_no_warmup: an on-demand GPU picture reader skips its warm-up encode

  #5698·Vop3p opened Nov 10, 2026Contributor·

  ·

• [Benchmark] RX 7900 XTX / Swift-1.5 IQ3_S / IQ3_S / strata-qwen38-thinking-v1-draft.1

  #5692·NemoNobody opened Nov 10, 2026·

  ·
  5

• [Engine] opt-in STRATA_VRAM_CALIB: size each card's VRAM reserve from what this binary and config were measured to allocate

  #5691·Vop3p opened Nov 9, 2026Contributor·

  ·

• [Benchmark] add synthetic two-request reproduction for shared Qwen repetition (#606)

  #5690·midhatn opened Nov 9, 2026Contributor·

  ·

• [Engine] preserve acknowledged thinking-budget drain tokens

  #5689·midhatn opened Nov 9, 2026Contributor·

  ·
  2

• [Benchmark] RX 7900 XTX / Qwen3.8-Flash-Next IQ3_S / Linux HIP gfx1100 community run

  #5688·str0g opened Nov 9, 2026·

  ·