Issues

• Allow system ram to be used instead of SSD when available.

  Niko1221/Strata#847 · Filter by author jarrodhroberson opened on Oct 4, 2026
  2

• Tool calls silently dropped in agent sessions after an empty assistant turn (fix: skip empty assistant turns when rendering)

  Niko1221/Strata#843 · Filter by author jramirez-uab opened on Oct 4, 2026

• cache files problem and other one problem

  Niko1221/Strata#842 · Filter by author gitforestit opened on Oct 4, 2026

• not stable start with 3+ cards

  Niko1221/Strata#841 · Filter by author yotadrivers-blip opened on Oct 4, 2026
  1

• --calibrate on Windows + AMD (HIP prebuilt 0.1.39) measures decode ~18x slower than the same engine through server.py, so the tuning is meaningless

  Niko1221/Strata#840 · Filter by author The-Dude-2020 opened on Oct 4, 2026

• Milestone: Add Apple Silicon Cross-Platform Support

  Niko1221/Strata#836 · Filter by author Atlessc opened on Oct 4, 2026

• The profile's length caps the expert arena — even an explicit --expert-cache N

  Niko1221/Strata#831 · Filter by author ZhongUncle opened on Oct 4, 2026

• Short one-shot prompts: a ~250 ms floor on the batched read, and ~45 ms of prompt-cache work that is never reused

  Niko1221/Strata#830 · Filter by author brenoperucchi opened on Oct 4, 2026

• Heap corruption (munmap_chunk: invalid pointer) at startup in Prefill::init — crashes with 1/2/4 GPUs

  Niko1221/Strata#828 · Filter by author wehooper4 opened on Oct 4, 2026
  1

• Results for --model UD-IQ4_XS

  Niko1221/Strata#825 · Filter by author Almazick opened on Oct 4, 2026
  2

• ROCm on Windows ZIP

  Niko1221/Strata#821 · Filter by author noctrex opened on Oct 4, 2026

• Assistant turn can be interrupted when generation emits a reserved ChatML boundary marker

  Niko1221/Strata#817 · Filter by author d4mer opened on Oct 4, 2026

• 0.1.39: decode on an RX 6800 (gfx1030, Windows) is 12-22% slower than 0.1.38; STRATA_SH_STREAM=0 removes it

  Niko1221/Strata#816 · Filter by author did-technomancer opened on Oct 4, 2026
  3

• Would it be possible to add support for the AMD Radeon RX 6750 GRE 12GB?

  Niko1221/Strata#814 · Filter by author lzb379440026 opened on Oct 4, 2026

• Strange behavior for x2 RTX 5060 Ti (16Gb each) and UD-Q4_K_XL

  Niko1221/Strata#806 · Filter by author art-den opened on Oct 4, 2026

• Tool call written inside <think> (no </think>) is returned as reasoning_content: no tool_calls, finish_reason=stop

  Niko1221/Strata#804 · Filter by author talisp opened on Oct 4, 2026

• Perplexity ~7-10% above llama.cpp on the same GGUF (Qwen3.8-Flash-Next), also at 2K context, every version tested

  Niko1221/Strata#803 · Filter by author mr20399 opened on Oct 4, 2026

• Hybrid CPU detection on Linux counts only the Turbo Boost Max 3.0 "favored" cores as P-cores (Core Ultra 7 270K Plus: 2P/22E instead of 8P/16E) — affects setup's --pool-workers and --pool-affinity auto|p-cores

  Niko1221/Strata#798 · Filter by author WUZICANGJIE opened on Oct 4, 2026

• Main engine strata.exe crashes with STATUS_ILLEGAL_INSTRUCTION (0xC000001D) on non-AVX-512 CPU - 20 WER crashes across v0.1.35/v0.1.38, fault offsets cluster in fixed code regions

  Niko1221/Strata#795 · Filter by author bjfr123321 opened on Oct 4, 2026
  1

• Community benchmark: IQ3_S and community AP-Q4_K_M — single RTX 5070 Ti vs 2× RTX 5060 Ti layer split

  Niko1221/Strata#791 · Filter by author yy16432 opened on Oct 4, 2026
  1

• Codex got error "input items of type 'additional_tools' are not supported"

  Niko1221/Strata#782 · Filter by author Anivie opened on Oct 4, 2026

• 1M context: identical expert work per window, but the VRAM-touching stages grow 7-8x

  Niko1221/Strata#781 · Filter by author 1314521gjy opened on Oct 4, 2026
  4

• laptop crash with latest

  Niko1221/Strata#779 · Filter by author XeonG opened on Oct 4, 2026
  1

• --batch on a 100%-resident layer split: engine dies at "captured the batch window over slots 0,1" (zero-doorbell path?)

  Niko1221/Strata#776 · Filter by author Seen-Tomorrow opened on Oct 4, 2026
  3

• Community result: 2.5-3.1x output at 256K context on a 16 GB card (IQ3_S) - calibration, --spec 6, learned profile

  Niko1221/Strata#775 · Filter by author JiuYue0820 opened on Oct 4, 2026
  1