Issues

• [Benchmark] RTX 3090 Ti, 24 GiB VRAM, 32GB RAM, qwen3.8-flash-next-iq2_xs

  Niko1221/Strata#3762 · Filter by author AlBundy33 opened on Oct 22, 2026

• [Engine] Grammar-constrained decoding for json_schema, so an agent's answers always fit its schema

  Niko1221/Strata#3758 · Filter by author qiuzan2001 opened on Oct 22, 2026

• Results for 128GB DDR5 RTX5080 16GB AMD Ryzen 9 9950X3D 16-Core Processor 2TB SSD PCIe 5.0 x4

  needs-format
  Niko1221/Strata#3755 · Filter by author DenisZDV opened on Oct 22, 2026
  1

• [GUI] an existing chat does not restore its sidebar workspace after restart (must re-select it to continue)

  Niko1221/Strata#3754 · Filter by author xvcezrqpbxtpzblrt opened on Oct 22, 2026

• [TUI] connect pi: compaction occurs at 15,000 tokens regardless of context window size

  Niko1221/Strata#3753 · Filter by author dnnkeeper opened on Oct 22, 2026

• [GUI] UI resource profile hard-codes vram_reserve_mib 3256: 256K context + Q4 KV leaves 0 MiB for the expert cache -> EngineDied at verify

  Niko1221/Strata#3750 · Filter by author il-freewind opened on Oct 22, 2026
  4

• [Other] Strata folder inconsistencies on non-english systems

  Niko1221/Strata#3749 · Filter by author Leseratte10 opened on Oct 22, 2026
  2

• [GUI] pasting a screenshot with Ctrl+V does nothing in the message box

  Niko1221/Strata#3748 · Filter by author xvcezrqpbxtpzblrt opened on Oct 22, 2026

• [Benchmark] RTX 3090 Ti, 24 GiB VRAM, 32GB RAM qwen3.8-flash-next-iq3_xxs

  Niko1221/Strata#3746 · Filter by author AlBundy33 opened on Oct 22, 2026

• Context cannot reset issues.

  needs-format
  Niko1221/Strata#3745 · Filter by author viczhangge opened on Oct 22, 2026
  2

• [Engine] session files (SAVE/RESTORE) refuse multi-GPU layer-split runs — request per-device session files

  Niko1221/Strata#3744 · Filter by author exotic16 opened on Oct 22, 2026

• [Engine] typed decisions: score a schema's token paths instead of generating them

  Niko1221/Strata#3743 · Filter by author leonardomurillo opened on Oct 22, 2026
  1

• [GUI] Models page shows the Starter (Qwen2.5-0.5B) for a few seconds on every launch

  Niko1221/Strata#3740 · Filter by author xvcezrqpbxtpzblrt opened on Oct 22, 2026
  1

• [Engine] Asymmetric GPUs and submodel pinning

  needs-format
  Niko1221/Strata#3739 · Filter by author ArnaudContet opened on Oct 22, 2026
  1

• [Engine] RFC: Extend NVFP4-QAD support to qwen35moe (Qwen 3.6 35B-A3B)

  Niko1221/Strata#3738 · Filter by author caioBB opened on Oct 22, 2026
  1

• [GUI] Can't put decimals into Expert mode Sampling Settings

  Niko1221/Strata#3736 · Filter by author therealMKD opened on Oct 22, 2026

• In-place GGUF expert pages stay GPU-pinned on top of the resident pool (~2x expert RAM); posix_fadvise(DONTNEED) frees 0 - blocks lend coverage (#669) on 128 GB boxes

  needs-format
  Niko1221/Strata#3734 · Filter by author bobgitmcgrath opened on Oct 22, 2026
  1

• K-quant experts (UD-Q5_K_XL) x peer GPU tier: HIP 719 launch failures + no-progress hangs on 50K+ token prompts (gfx1100 primary + gfx1151 peer, 0.1.47-dual)

  needs-format
  Niko1221/Strata#3733 · Filter by author bobgitmcgrath opened on Oct 22, 2026
  1

• [Engine] Engine crashes unexpectedly (exit code 1 / 3221226505) during DeepSeek Harness session on Windows 11

  Niko1221/Strata#3732 · Filter by author Gelomen opened on Oct 22, 2026

• [GUI] AI reply cannot be copied in the desktop app (Copy response does nothing, no feedback)

  Niko1221/Strata#3731 · Filter by author xvcezrqpbxtpzblrt opened on Oct 22, 2026

• [Engine] Per-pack MTP draft head (or q8/q4 MTP expert runtime) for non-Q2_0 packs

  needs-format
  Niko1221/Strata#3728 · Filter by author dduckpal opened on Oct 22, 2026
  1

• [GUI] Windows + AMD (RX 9070 XT): GPU load/VRAM/temp/power/PCIe "Not available", onboarding shows 0.0 GB VRAM

  Niko1221/Strata#3727 · Filter by author LausitzAir opened on Oct 22, 2026
  3

• [Setup] Low-RAM mode: KV streaming is never chosen, though it frees VRAM for experts at the cost of ~1.7 GB of RAM (patch inside)

  needs-format
  Niko1221/Strata#3726 · Filter by author altiber opened on Oct 22, 2026
  1

• [Engine] Windows: the auto RAM complement ignores the WDDM shared-segment cap, the engine dies at start on a 2-GPU layer split (patch inside)

  Niko1221/Strata#3725 · Filter by author altiber opened on Oct 22, 2026

• [Benchmark] llama.cpp #30553 improves IQ3_S throughput on RTX 2080 Ti + RTX 4070 SUPER

  Niko1221/Strata#3724 · Filter by author Chenj168 opened on Oct 22, 2026