Issues

• [Other] IQ3_S install stops in the MTP draft-layer fetch (57% of gate_up_proj) and every retry fails; the host's own fetch ignores the Windows proxy

  Niko1221/Strata#4305 · Filter by author redfireblade opened on Oct 27, 2026

• [Engine] 0.1.51 CUDA sm_86 (RTX 3090, UD-Q4_K_XL): after 2 days of serving, every prompt ≥20K tokens ends non-finite (#879) until a restart — NaN in the 5-token short-read tail

  Niko1221/Strata#4304 · Filter by author miomao656 opened on Oct 27, 2026

• [GUI] Model list shows DeepSeek-V4.1-Flash under "Qwen3.6 models" group in 0.2.6

  Niko1221/Strata#4303 · Filter by author Jacky-Yan040212 opened on Oct 27, 2026

• Strata 0.2.6 · Engine 0.1.50 Messed up toolcalls

  needs-format
  Niko1221/Strata#4299 · Filter by author XeonG opened on Oct 27, 2026
  1

• [GUI] Monitor: show cached, reading and generated tokens in the Context card

  Niko1221/Strata#4297 · Filter by author thisisspartan opened on Oct 27, 2026

• Tecnoter Group: Compressori e Generatori a Motore Idraulico per Mezzi di Servizio Sempre Operativi

  needs-format
  Niko1221/Strata#4294 · Filter by author tecnotergroup2-a11y opened on Oct 27, 2026
  1

• [Engine] V100 (sm_70) 2-card layer split: engine 0.1.51 reads a 32K prompt ~5% slower than 0.1.43.1 (server ruled out by a crossed test)

  Niko1221/Strata#4292 · Filter by author zhaochengggg opened on Oct 27, 2026

• [GUI] Add bidirectional copy-and-move functionality for dual-panel files and batch file renaming [Feature].

  Niko1221/Strata#4291 · Filter by author hermitguo opened on Oct 27, 2026
  2

• [Benchmark] RTX 4080 (sm_89): sm_86 switch set gives +7% agent decode

  Niko1221/Strata#4290 · Filter by author Fuyyzlogik opened on Oct 27, 2026
  2

• [GUI] Monitor: /v1 proxy requests show no VRAM hit rate, and native engine metrics are not shown

  Niko1221/Strata#4289 · Filter by author Songokou1983 opened on Oct 27, 2026

• [Benchmark] Qwen3.8-Flash-Next on AMD Ryzen AI 9 365 (32GB Unified Memory)【硬核整活】Ryzen AI 9 365 + 880M核显，我用32G统一内存硬啃125B巨无霸MoE大模型！

  Niko1221/Strata#4288 · Filter by author itandpcjob opened on Oct 27, 2026
  2

• 是否支持A800 A100 H100 这样的显卡

  needs-format
  Niko1221/Strata#4287 · Filter by author xxch opened on Oct 27, 2026
  1

• [Benchmark] 4x V100-SXM2 32 GB (NVLink) + RTX 3090 Ti, Qwen3.8-Flash-Next NVFP4 layer split on 0.2.6: 8 x 262K slots, 2 x 524K, needle 3/3 to 500K

  Niko1221/Strata#4285 · Filter by author andyelka-creator opened on Oct 27, 2026
  3

• [Engine] Deadlock on ~25K-token context: /v1/chat/completions hangs forever, engine busy-loops ~350% CPU, /v1/models stays 'loaded' (0.1.48, RTX 5080, Linux)

  Niko1221/Strata#4284 · Filter by author dedvmedved-dot opened on Oct 27, 2026

• [Engine] Opt-in routed prompt stream: do not copy the streamed experts no token of the chunk routes to

  Niko1221/Strata#4282 · Filter by author bitaccesscomau opened on Oct 27, 2026

• [Benchmark] Community: Qwen3.8-Flash-Next IQ3_S + DeepSeek-V4.1-Flash IQ2_XXS on one RTX 4000 Ada 20 GB (Windows)

  Niko1221/Strata#4281 · Filter by author pietroMastro92 opened on Oct 27, 2026

• [Other] Cannot load models

  Niko1221/Strata#4280 · Filter by author elppato opened on Oct 27, 2026
  2

• [Engine] Strata update downloads CUDA engine instead of HIP on Radeon AI PRO R9700 (0.2.6)

  Niko1221/Strata#4279 · Filter by author ecopspawel opened on Oct 27, 2026

• In opencode, context compression cannot be invoked automatically.

  needs-format
  Niko1221/Strata#4278 · Filter by author bai-yongfei opened on Oct 27, 2026
  2

• is it surpport REDHAT 7.6 SYSTEM?my Nvidia driver latest version is 535

  needs-format
  Niko1221/Strata#4276 · Filter by author yanghp86 opened on Oct 27, 2026
  1

• [FEATURE REQUEST] - GPU unlocks?

  needs-format
  Niko1221/Strata#4275 · Filter by author SabinStargem opened on Oct 27, 2026
  1

• [Engine] ship a cuda12 strata-dsv41 package (sm_70, sm_75) so Volta and Turing can run DeepSeek-V4.1-Flash

  Niko1221/Strata#4273 · Filter by author yudagua opened on Oct 27, 2026
  2

• [Engine] serve: a client that stops reading can block the engine for everyone, and waiting requests are not served in arrival order

  Niko1221/Strata#4271 · Filter by author MikeLippold opened on Oct 26, 2026

• AMD MI100 performance drop after 0.1.40.3 on 0.2.4 and 0.2.6

  needs-format
  Niko1221/Strata#4269 · Filter by author yoyz opened on Oct 26, 2026
  1

• [Engine] TQ MMQ overlay fails to build against current upstream GGML (loader x_gate parameter)

  Niko1221/Strata#4266 · Filter by author Tsury opened on Oct 26, 2026