Issues

• proposal(ascend): experimental single-card dense GGUF backend

  kkokosa/dotLLM#447 · Filter by author Cheongroach opened on Oct 28, 2026

• server(perf): reduce stop-list allocations during request conversion

  kkokosa/dotLLM#443 · Filter by author Cheongroach opened on Oct 24, 2026

• server(api): reject unsupported chat choice counts and preserve omitted n

  kkokosa/dotLLM#442 · Filter by author Cheongroach opened on Oct 24, 2026

• engine(kv-cache): reject invalid SimpleKvCache dimensions before allocation

  kkokosa/dotLLM#441 · Filter by author Cheongroach opened on Oct 24, 2026

• cpu(threading): DecodeThreadCount=1 is silently raised to two threads

  kkokosa/dotLLM#439 · Filter by author Cheongroach opened on Oct 24, 2026

• server: absolute GGUF paths include the extension in ModelId

  kkokosa/dotLLM#437 · Filter by author Cheongroach opened on Oct 18, 2026

• docs(kv-cache): correct memory sizing, CPU layout, and decode complexity

  kkokosa/dotLLM#435 · Filter by author Cheongroach opened on Oct 18, 2026

• bug(cuda): int overflow in weight-upload/dequant byte-size products for >2 GiB tensors

  kkokosa/dotLLM#432 · Filter by author jamesburton opened on Aug 9, 2026

• bug(cpu/matmul): int overflow in Q8_0/F16 GEMV weight row offsets for >2 GB tensors

  kkokosa/dotLLM#429 · Filter by author unsafePtr opened on Aug 4, 2026
  1

• docs(server): lora_adapter is documented as a request field but implemented nowhere

  kkokosa/dotLLM#425 · Filter by author jamesburton opened on Jul 31, 2026
  1

• engine(streaming): token-id callback cannot express stop-string-trimmed output

  kkokosa/dotLLM#424 · Filter by author jamesburton opened on Jul 31, 2026

• engine(samplers): top-p tie-breaking is undefined at large vocab (float ties + unstable sort)

  kkokosa/dotLLM#423 · Filter by author jamesburton opened on Jul 31, 2026

• server(sse): Connection: keep-alive is sent on the OpenAI streaming endpoints (prohibited over HTTP/2+)

  kkokosa/dotLLM#422 · Filter by author jamesburton opened on Jul 31, 2026

• test(unit): ~26 hardware-gated tests report as passed instead of skipped

  kkokosa/dotLLM#421 · Filter by author jamesburton opened on Jul 31, 2026

• perf(cpu): AVX2 beats both 512-bit tiers in the Q8_0 row-major dot product, but ComputeRows prefers them

  kkokosa/dotLLM#415 · Filter by author kkokosa opened on Jul 30, 2026
  2

• perf(tokenizers): BPE encoder allocates ~32 bytes per input character

  enhancement
  kkokosa/dotLLM#413 · Filter by author kkokosa opened on Jul 29, 2026
  1

• ComputeThreadPool workers idle threads: ceiling-division partitioning leaves up to ~47% of the pool unused

  kkokosa/dotLLM#402 · Filter by author kkokosa opened on Jul 29, 2026

• Drop the row-major weight duplicate for Q8_0 (blocked on fused decode still reading it)

  kkokosa/dotLLM#401 · Filter by author kkokosa opened on Jul 29, 2026
  1

• R4 repacking: extend to prefill via interleaved GEMM, and fix ~2x under-reported memory

  kkokosa/dotLLM#397 · Filter by author kkokosa opened on Jul 28, 2026
  1

• perf(vulkan): MoE indexed-expert prefill matmul is scalar, no tiling/dp4a -- 2.3% of llama.cpp on Qwen3.6-35B-A3B

  kkokosa/dotLLM#383 · Filter by author jamesburton opened on Jul 18, 2026
  3

• perf(vulkan): adaptive S/M/L coopmat warptile sizing + split_k for prefill GEMM (close 0.65x SmolLM prefill gap)

  kkokosa/dotLLM#375 · Filter by author jamesburton opened on Jul 17, 2026
  1

• bug(cpu): DequantizeQ5_KAvx2 throws AccessViolationException at large tensor sizes

  kkokosa/dotLLM#374 · Filter by author jamesburton opened on Jul 17, 2026

• GGUF v2 files fail to load: counts/lengths parsed as uint32 instead of uint64

  kkokosa/dotLLM#364 · Filter by author jamesburton opened on Jun 30, 2026

• perf(cuda): coalesced Q4_K MMVQ decode GEMV for Turing/T4 (~2× decode target)

  kkokosa/dotLLM#363 · Filter by author jamesburton opened on Jun 29, 2026

• scheduler: cross-device KV handoff for DisaggregatedScheduler (device→host→device + per-replica device selection)

  kkokosa/dotLLM#360 · Filter by author jamesburton opened on Jun 29, 2026