Issues

• Qwen3-TTS on NPU

  peonist-ai/halogen-flash-server#183 · Filter by author 0x62ash opened on Oct 14, 2026
  5

• WTF thinking loop

  peonist-ai/halogen-flash-server#180 · Filter by author 0xmarcinz opened on Oct 14, 2026
  6

• [Feature request]: image edits with a reference image (compose an asset into a target picture)

  peonist-ai/halogen-flash-server#179 · Filter by author leng45-cpu opened on Oct 14, 2026
  1

• 0.18.1 GGUF trunk: concurrent requests intermittently return an empty reply (first token is EOS); solo identical prompts always correct

  peonist-ai/halogen-flash-server#172 · Filter by author JohannesBertens opened on Oct 12, 2026
  6

• [Feature Request] Explore Qwen-Image-2.1-Turbo on NPU with INT8 and a lightweight Zen-style encoder

  peonist-ai/halogen-flash-server#164 · Filter by author kamjin3086 opened on Oct 11, 2026

• Concurrent long prefill stalls in-flight decodes (ITL p90 49 ms → 10.6 s)

  peonist-ai/halogen-flash-server#161 · Filter by author dardeaw opened on Oct 11, 2026
  4

• Feature Proposal: Explore CPU Inference Alongside GPU and NPU Models on Strix Halo

  peonist-ai/halogen-flash-server#159 · Filter by author xuyongroger opened on Oct 10, 2026
  2

• crash logged with system+user+user payload

  peonist-ai/halogen-flash-server#146 · Filter by author disarticulate opened on Oct 7, 2026
  10

• What precision is the KV cache stored in, and is KV quantization supported?

  peonist-ai/halogen-flash-server#142 · Filter by author apartje opened on Oct 6, 2026
  8

• Strix Halo hard hangs from NPU+GPU overlap with fabric clock held — sustained and brief (0.16.1/0.16.2)

  peonist-ai/halogen-flash-server#139 · Filter by author Thaeland opened on Oct 5, 2026
  2

• Publish a build workflow with provenance attestation for the GHCR images

  peonist-ai/halogen-flash-server#136 · Filter by author hjanuschka opened on Oct 4, 2026

• Paired PPL: unsloth UD-Q4_K_XL is ~2% lower than v2 on EN/ZH prose (tie on code) at ~30% lower decode; is a higher-precision trunk planned?

  enhancement
  peonist-ai/halogen-flash-server#130 · Filter by author zengjiajun0623 opened on Oct 3, 2026
  5

  

• [Feature]: Support REAP models

  enhancement
  peonist-ai/halogen-flash-server#103 · Filter by author MrDrMcCoy opened on Sep 23, 2026
  6

  

• The shipped checkpoint predicts <|im_end|> mid-prose past ~96k tokens, and the sidecar makes it worse

  peonist-ai/halogen-flash-server#89 · Filter by author wenis opened on Sep 21, 2026
  12

• Tracking: stalls and wedges under host memory pressure (#71, #83, #79)

  peonist-ai/halogen-flash-server#85 · Filter by author wenis opened on Sep 20, 2026
  7

• Persist composable-context chunks to HALOGEN_CACHE_DIR

  enhancement
  peonist-ai/halogen-flash-server#70 · Filter by author lev-medien-sandkasten opened on Sep 17, 2026
  1

  

• Prefix caching could be improved.

  enhancement
  peonist-ai/halogen-flash-server#31 · Filter by author carrot-root-ai opened on Sep 12, 2026
  11