Issues

• Paiged Attention

  gufo-org/gufo#873 · Filter by author zugliebhaber opened on Oct 29, 2026

• perf(deepseek): 5.5%-18% improvements in decode and 5.9% faster prefill

  gufo-org/gufo#867 · Filter by author sdwilsh opened on Oct 29, 2026
  5

• CLI + Lemonade Bridge

  gufo-org/gufo#865 · Filter by author perl5punk opened on Oct 28, 2026

• Flash-Next MTP decode (tg128) dropped ~12.6% between main 8c1eeb98 (ROCm 7.2.3) and 78d74185 (ROCm 10.1)

  gufo-org/gufo#863 · Filter by author francescobozzo opened on Oct 28, 2026
  3

• Field report: 0.17 (no disk cache) far more stable than 0.24 for long agentic sessions on Strix Halo

  gufo-org/gufo#853 · Filter by author kvzp opened on Oct 27, 2026
  2

• Would you be interested in MOSS-Image (OpenMOSS, Apache-2.0) text-to-image?

  Low-prioFilter by Priority: Low-prio

  High-effortFilter by Effort: High-effort
  Feature

  gufo-org/gufo#852 · Filter by author D3rPapts opened on Oct 27, 2026
  2

  

• Support gfx1152 (Kraken Point)

  Medium-prioFilter by Priority: Medium-prio

  Low-effortFilter by Effort: Low-effort
  Feature

  gufo-org/gufo#843 · Filter by author fkz opened on Oct 27, 2026
  1

  

• TTS: Native gfx1151 support for Fish Audio Drama-3 (S2-Pro 30B-A3B MoE + Gated DeltaNet, 44.1 kHz)

  Feature

  gufo-org/gufo#838 · Filter by author rikaaa0928 opened on Oct 27, 2026
  

• Windows: loaders that copy weights through the file mapping keep the whole model in host RAM (Qwen-Image 2.1 measured, MiniMax H3 likely)

  gufo-org/gufo#829 · Filter by author D3rPapts opened on Oct 26, 2026
  1

• Benchmarks: comparable community results (machine identity, a place to submit, practical workloads)

  gufo-org/gufo#809 · Filter by author D3rPapts opened on Oct 26, 2026

• image: LoRA adapters for Qwen-Image-2.1 (merged at upload, startup then per-request)

  Qwen image 2.1Filter by Model: Qwen image 2.1

  Low-prioFilter by Priority: Low-prio

  High-effortFilter by Effort: High-effort
  Feature

  gufo-org/gufo#797 · Filter by author naerymdan opened on Oct 25, 2026

• cache: A fixed 4 GiB heap reserve keeps the RAM cache off when GTT is tight

  gufo-org/gufo#781 · Filter by author sdwilsh opened on Oct 24, 2026

• Qwen3.8-27B DFlash2: sampled answers that repeat themselves lose ~7 % with #705's prompt-lookup gate

  Qwen3.8 27bFilter by Model: Qwen3.8 27b

  Medium-prioFilter by Priority: Medium-prio

  High-effortFilter by Effort: High-effort
  Enhancement

  gufo-org/gufo#771 · Filter by author D3rPapts opened on Oct 24, 2026
  2

• embeddings: complete EmbeddingGemma 2 processor controls

  EmbeddingGemma 2Filter by Model: EmbeddingGemma 2

  Medium-prioFilter by Priority: Medium-prio

  Low-effortFilter by Effort: Low-effort
  Feature

  gufo-org/gufo#763 · Filter by author fedeizzo opened on Oct 24, 2026
  2

• embeddings: complete EmbeddingGemma 2 support

  EmbeddingGemma 2Filter by Model: EmbeddingGemma 2

  Medium-prioFilter by Priority: Medium-prio

  Low-effortFilter by Effort: Low-effort
  Feature

  gufo-org/gufo#760 · Filter by author fedeizzo opened on Oct 24, 2026

• Long-running request hangs for ~2 hours, then returns stream_errorwith deadline_exceeded; client (opencode) shows "interrupted" and stops

  Medium-prioFilter by Priority: Medium-prio

  High-effortFilter by Effort: High-effort
  Bug

  gufo-org/gufo#753 · Filter by author uiice opened on Oct 24, 2026

• perf(qwen38_flash_next): #699 moves ~7-10 ms of first-use setup into the first MTP decode on Windows/TheRock

  Qwen3.8 Flash NextFilter by Model: Qwen3.8 Flash Next

  High-prioFilter by Priority: High-prio

  Medium-effortFilter by Effort: Medium-effort
  Enhancement

  gufo-org/gufo#745 · Filter by author thomas9120 opened on Oct 23, 2026
  1

  

• Feature request: EXL3 weights for Qwen3.8-Flash-Next (same quality as UD-Q4_K_XL, ~7 GB smaller)

  Qwen3.8 Flash NextFilter by Model: Qwen3.8 Flash Next

  Low-prioFilter by Priority: Low-prio

  High-effortFilter by Effort: High-effort
  Feature

  gufo-org/gufo#737 · Filter by author baldlawyer opened on Oct 23, 2026
  

• serve: Unknown request fields return 400

  Medium-prioFilter by Priority: Medium-prio

  Low-effortFilter by Effort: Low-effort
  Enhancement

  gufo-org/gufo#724 · Filter by author sdwilsh opened on Oct 23, 2026

• About version show

  gufo-org/gufo#721 · Filter by author uiice opened on Oct 23, 2026
  4

  

• CPU hot paths: runtime-dispatched AVX-512 and a division-free n-gram hash

  Qwen3.8 Flash NextFilter by Model: Qwen3.8 Flash Next

  Medium-prioFilter by Priority: Medium-prio

  Medium-effortFilter by Effort: Medium-effort
  Enhancement

  gufo-org/gufo#716 · Filter by author winaes13-ux opened on Oct 22, 2026

• serve: opt-in overcommit of session state (grow on demand) so 4x262144 fits where 2x262144 queues

  Qwen3.8 Flash NextFilter by Model: Qwen3.8 Flash Next

  Medium-prioFilter by Priority: Medium-prio

  Medium-effortFilter by Effort: Medium-effort
  Bug

  gufo-org/gufo#715 · Filter by author philtheriver opened on Oct 22, 2026

• Feature request: Optional strict protocol handling with corrective regeneration

  Low-prioFilter by Priority: Low-prio

  High-effortFilter by Effort: High-effort
  Feature

  gufo-org/gufo#706 · Filter by author still-dreaming-1 opened on Oct 22, 2026

• Add support for Qwen Image 2.1 Turbo

  Qwen image 2.1Filter by Model: Qwen image 2.1

  Low-prioFilter by Priority: Low-prio

  High-effortFilter by Effort: High-effort
  Feature

  gufo-org/gufo#689 · Filter by author e270889o opened on Oct 20, 2026

• Round-robin prefill turns (#470) delay first tokens and discard the n-gram prefetch (#683) for concurrent long prompts

  serving

  Medium-prioFilter by Priority: Medium-prio

  Medium-effortFilter by Effort: Medium-effort
  Enhancement

  gufo-org/gufo#687 · Filter by author neuhaus opened on Oct 20, 2026