Issues

• [Feature]: --kv-grow with --prefill auto: let the prompt path's loan shrink as the K/V grows (auto stops a 262K config at ~115K on a 12 GB card)

  Niko1221/Strata#2310 · Filter by author JawadN00r opened on Oct 13, 2026

• prefill copy_i32: misaligned address

  Niko1221/Strata#2309 · Filter by author dandandelion opened on Oct 13, 2026

• When I have any conversation, it will be cancelled directly

  Niko1221/Strata#2307 · Filter by author zhaojigang opened on Oct 13, 2026
  1

• web app: the default temperature is 0.6; the Qwen3.8 model card and generation_config.json say 1.0

  Niko1221/Strata#2306 · Filter by author i-Vladimir-i opened on Oct 13, 2026

• HIP (R9700, gfx1201): the MTP draft is fast or slow depending on the engine start; GPU_MAX_HW_QUEUES=2 makes every start the fast one

  Niko1221/Strata#2305 · Filter by author i-Vladimir-i opened on Oct 13, 2026

• calibrate: configured GPU vision encoder is absent during auto expert-cache sizing

  Niko1221/Strata#2301 · Filter by author Tsury opened on Oct 13, 2026

• Agent history with long tool_calls arguments produces reasoning_content + empty content on final turns

  Niko1221/Strata#2297 · Filter by author bys13 opened on Oct 13, 2026

• 16GB RTX 5080 + 96GB RAM: intermittent VirtualAlloc error 1450 on expert-arena large pages + long-context (150-200k) stability data

  Niko1221/Strata#2295 · Filter by author guodongliang5689 opened on Oct 13, 2026

• Community benchmark on NVIDIA GeForce RTX 4080 SUPER

  Niko1221/Strata#2293 · Filter by author johlim opened on Oct 13, 2026

• How do you specify the thinking mode?

  Niko1221/Strata#2292 · Filter by author Coderx7 opened on Oct 13, 2026

• Community benchmark on Quadro P6000 - swift-1-5-iq3-xxs

  Niko1221/Strata#2291 · Filter by author almirjakubaitis opened on Oct 13, 2026

• Anthropic API: cache_creation_input_tokens is never reported, though every newly read token is kept

  Niko1221/Strata#2289 · Filter by author xrip opened on Oct 13, 2026

• [Bug] SYCL on a Meteor Lake iGPU (8086:7d55) won't start, then freezes: 3 small fixes resolves issues

  Niko1221/Strata#2282 · Filter by author Bassai-Sho opened on Oct 13, 2026

• [0.1.43.1] 2× RTX 3090 CUDA layer split: decode −20.5% vs v0.1.43, and STRATA_ADAPT_OVERLAP=1 cannot re-enable the overlap ('off: not on a layer split')

  Niko1221/Strata#2281 · Filter by author ben0112 opened on Oct 13, 2026

• Bug Description

  Niko1221/Strata#2279 · Filter by author eewey777 opened on Oct 13, 2026
  1

• Feature: legacy completions on /v1/completions (raw-prompt clients)

  Niko1221/Strata#2277 · Filter by author kossum opened on Oct 13, 2026

• [Direction] Aim Strata at long-context multi-user: a peer's prompt read leaves a decoding slot at 3.67 tok/s, the batch window emits 1 token/slot, and 4 streams take 4.04x a single read to their first tokens (2x2080 Ti layer split, 96K)

  Niko1221/Strata#2275 · Filter by author superniker opened on Oct 13, 2026

• # Community benchmark: RX 7900 XTX 24 GB on Windows, i7-12700K, 128 GB DDR4 — IQ3_S

  Niko1221/Strata#2274 · Filter by author wburzyns opened on Oct 13, 2026
  1

• Dell 7920超过9分钟的启动bug

  Niko1221/Strata#2269 · Filter by author tacom1 opened on Oct 13, 2026

• [0.1.43.1] Blackwell sm_120 (RTX PRO 6000): first CUDA-side measures of the opt-in switch season - GDN_CHUNKED and the S25/S26 fused stack measure neutral on the fully-resident tier; greedy outputs drift on 2/12 probes under the stack

  Niko1221/Strata#2268 · Filter by author beastrobin opened on Oct 13, 2026
  1

• iq3-xss engine keeps stopping with code -6

  Niko1221/Strata#2267 · Filter by author lama7 opened on Oct 13, 2026
  1

• Marker-shaped text truncates agent turns (EOS sampling + semantic priming)

  Niko1221/Strata#2266 · Filter by author nashcap opened on Oct 13, 2026

• Conversation cache re-parks the same conversation every turn: 2251.9 GB of host copies at 4.7 % reuse, and the K/V retention that would make the write incremental can never apply

  Niko1221/Strata#2264 · Filter by author Krypto-Whitehat opened on Oct 13, 2026

• Community benchmark on NVIDIA GeForce RTX 4090

  Niko1221/Strata#2263 · Filter by author street-tripy opened on Oct 13, 2026

• Community benchmark: RTX 5080 + RTX 5060 Ti 16 GB hold layers at 262K on Unsloth UD-Q4_K_XL — engine 0.1.43, two launch configs (fork-free submission, files attached)

  Niko1221/Strata#2262 · Filter by author beplee opened on Oct 13, 2026