All pull requests

Pull requests

• Prompt chunks sized for a layer split's pipeline (STRATA_PREFILL_PIPE=1, opt-in)

  #1441·ruibeikaa opened Oct 7, 2026Contributor·

  ·

• docs: report RTX 8000 measurements of existing runtime switches

  #1435·agorevski opened Oct 7, 2026·

  ·

• perf(peer): add opt-in byte-capacity-aware expert placement

  #1438·agorevski opened Oct 7, 2026·

  ·

• fix(serve): isolate CPU cores for independent Linux replicas

  #1436·agorevski opened Oct 7, 2026·

  ·

• perf(prefill): cache immutable BF16-to-FP16 weight conversions

  #1439·agorevski opened Oct 7, 2026·

  ·

• perf(cuda): add opt-in shape-tuned sm75 interleaved verify

  #1437·agorevski opened Oct 7, 2026·

  ·

• telemetry: GPU load, VRAM, temperature and power for AMD cards on Windows (#1380)

  #1434·gt50 opened Oct 7, 2026·

  ·
  1

• community bench: RTX 5090 + RTX PRO 4000 Blackwell, UD-Q4_K_XL helper cache, engine 0.1.40.3

  #1433·sophieandreikina opened Oct 7, 2026·

  ·

• bench: Arc Pro B65 Gen4 results on Strata v0.1.40.2

  #1432·timnevits opened Oct 7, 2026Contributor·

  ·

• serve: a tool call written <function= NAME> is the call NAME

  #1430·itripn opened Oct 7, 2026·

  ·

• bench: 2x TITAN RTX (sm_75) at Strata v0.1.40.3, IQ3_S at 262K

  #1429·shrisha108 opened Oct 7, 2026Contributor·

  ·
  1

• web: Russian UI localization (dictionary + header language picker)

  #1427·Astemiir opened Oct 7, 2026·

  ·

• dflash: experimental standalone DeepSpec DFlash block drafter for Qwen3.8-Flash-Next (greedy, correctness-first)

  #1426·j-luwierski opened Oct 7, 2026Contributor·

  ·

• Pascal GP100: a Q8_0 decode GEMV (STRATA_Q8_SM60=1, opt-in) and IQ3_XXS tables in shared memory

  #1424·ruibeikaa opened Oct 7, 2026Contributor·

  ·

• docs/kubernetes: manifests for one Strata server on a GPU node (kubectl, Kustomize, kapp)

  #1421·blange48 opened Oct 7, 2026Contributor·

  ·

• Add optional GLM-5.3-Flash support from Project Maya

  #1420·apr3ndi5 opened Oct 7, 2026·

  ·
  1

• bench: replay token IDs and verified windows in serve decode

  #1419·imanu86 opened Oct 7, 2026Contributor·

  ·
  1

• cuda: opt-in MMVQ activation reuse across output rows

  #1418·imanu86 opened Oct 7, 2026Contributor·

  ·
  1

• prefill: scope each split-stage loan estimate to its device

  #1417·imanu86 opened Oct 7, 2026Contributor·

  ·
  1

• prefill: the CPU share's activations quantized on the GPU, copied down beside the routing sync (builds on #1414)

  #1416·architectds opened Oct 7, 2026Contributor·

  ·

• cpu: add bit-exact AVX2 singleton expert rows

  #1415·InB4DevOps opened Oct 7, 2026·

  ·

• prefill: the CPU share (STRATA_PREFILL_CPU_SHARE) up to 3,072-token chunks, from mapped experts, on a layer split

  #1414·architectds opened Oct 7, 2026Contributor·

  ·
  1

• OLDER_GPUS.md: the P100 row now has a measurement (2x P100, IQ3_S)

  #1411·lechlna000 opened Oct 7, 2026Contributor·

  ·

• docs: a short "Contributing a change or report" section in AGENTS.md and a test-requests list

  #1408·paulhothersall opened Oct 7, 2026·

  ·
  1

• bench: community report, RX 7900 XTX (gfx1100) on WSL2, Coder IQ1_M: 0.1.33 vs 0.1.40, prefill streaming, 16k-256k sweep

  #1406·theGiallo opened Oct 7, 2026·

  ·