Issues

• AMD GPU 9060 XT is not suppported

  Niko1221/Strata#3019 · Filter by author lumylovepandas opened on Oct 18, 2026

• qwen35 (Qwen3.6): --mtp N is silently inert in serve mode unless --batch N --batch-mtp are also passed; setup's own config omits them

  Niko1221/Strata#3009 · Filter by author FlyoverBlind opened on Oct 18, 2026

• qwen35 (Qwen3.6): --mtp N is silently inert in serve mode unless --batch N --batch-mtp are also passed; setup's own config omits them

  Niko1221/Strata#3008 · Filter by author FlyoverBlind opened on Oct 18, 2026

• 这个性能指标咋样

  Niko1221/Strata#3007 · Filter by author phuai007 opened on Oct 18, 2026
  1

• [Bug]: The Mac (Metal) engine ignores a request's sampling: temperature, top_p, top_k, min_p, seed and the penalties do nothing, logit_bias is dropped, and every answer is the greedy one

  Niko1221/Strata#3006 · Filter by author oceanio opened on Oct 18, 2026

• getting errors on 0.1.45 : registration error

  Niko1221/Strata#3005 · Filter by author Schumi08 opened on Oct 18, 2026

• bench: community report, 4x Tesla T4 (sm_75, Linux, Docker): UD-Q6_K_XL on 0.1.45, +17% decode vs 0.1.42, 128K context verified with a 93K prompt, and the per-request VRAM warning that --expert-cache auto sets off by itself

  Niko1221/Strata#2999 · Filter by author ericoh88 opened on Oct 18, 2026

• KV cache q8_0 + q5_1

  Niko1221/Strata#2992 · Filter by author adbensi opened on Oct 18, 2026

• [Bug]: ple_reader_selftest can crash or hang when two runs share the temp directory

  Niko1221/Strata#2991 · Filter by author thc1006 opened on Oct 18, 2026

• [Bug]: sycl/tools/build.sh exits 1 after a successful build and 0 after a failed compile

  Niko1221/Strata#2990 · Filter by author thc1006 opened on Oct 18, 2026

• 2026-10-17-community-rx-5500-xt-gfx1012 Bnechmarks and build information.

  Niko1221/Strata#2983 · Filter by author cryptofuse opened on Oct 17, 2026

• Qwen3.6-35B-A3B UD-Q4_K_XL: Low GPU utilization, limited generation speed, and MTP acceptance concerns

  Niko1221/Strata#2982 · Filter by author cinetube opened on Oct 17, 2026
  1

• serve: /v1/chat/completions ignores llama.cpp's n_predict, so a client's n_predict: 0 follow-up call regenerates the whole answer with no limit (~20k tokens, ~4–5 min of GPU time per reply)

  Niko1221/Strata#2980 · Filter by author DanGTG opened on Oct 17, 2026

• Layer split + --batch: slots decode without MTP drafts, and --batch-mtp is slower on 4 stages. Could pipelined groups draft?

  Niko1221/Strata#2973 · Filter by author JacobSussan opened on Oct 17, 2026

• Web app: page overflows sideways on phones when a tool call appears (.view's automatic minimum size)

  Niko1221/Strata#2971 · Filter by author TitanMamo opened on Oct 17, 2026
  3

• Expose optional pricing in model listings for client usage estimates

  Niko1221/Strata#2969 · Filter by author lindestad opened on Oct 17, 2026

• Windows HIP, 0.1.44.1, one RX 7900 XTX: about 1 start in 5 the KV streaming does no block reads, and every prompt past --kv-resident gets a wrong answer

  Niko1221/Strata#2960 · Filter by author jerem91150 opened on Oct 17, 2026

• 我发现pcie的速度对prefill影响很大

  Niko1221/Strata#2958 · Filter by author chanverlao opened on Oct 17, 2026
  2

• Community benchmark NVFP4 - RTX 5060 Ti 16GB - Ryzen 5600 - 96GB RAM + needle_bench (Qwen3.8-Flash-Next-NVFP4-GPTQ-Strata)

  Niko1221/Strata#2956 · Filter by author tugurufu-netizen opened on Oct 17, 2026

• [Bug]: default --pipeline-windows 2 costs ~25% decode on 2x R9700 with every expert resident (v0.1.45)

  Niko1221/Strata#2948 · Filter by author justincinmd opened on Oct 17, 2026

• [Feature]: Optional rootless Apple Silicon Monitor sensors

  Niko1221/Strata#2942 · Filter by author mjhagen opened on Oct 17, 2026

• 0.1.45: non-thinking requests inherit the card's presence_penalty 1.5 (temperature 0 too), ~25% slower decode

  Niko1221/Strata#2940 · Filter by author Cass67 opened on Oct 17, 2026
  1

• 4× Tesla T10 (Turing, layer split): single-stream decode stays ~39 tok/s with 99.9% of experts in VRAM — same as 2× T10

  Niko1221/Strata#2939 · Filter by author demon851113 opened on Oct 17, 2026
  1

• Community benchmark: RX 7900 XTX 24 GB (gfx1100), Ryzen 9 9950X, 96 GB DDR5-6400 — 0.1.45 HIP source build, IQ3_S, 262K

  Niko1221/Strata#2938 · Filter by author destroyor opened on Oct 17, 2026

• Metal build broken on current main (v0.1.45-metal): 3 separate defects in the perf-merge sync

  Niko1221/Strata#2937 · Filter by author DarkJoll opened on Oct 17, 2026