All pull requests

Pull requests

• feat(server)!: add parallel tool-call controls

  #1603·RemiliaForever (RemiliaForever) opened Oct 26, 2026Contributor·11 of 12 tasks·

  ·
  1

• feat(sdk): build llama.cpp with gemma4-htp-opt fork patch

  #1602·Mengsheng Wu (mengshengwu) opened Oct 26, 2026Contributor·3 tasks done·

  ·
  

• ci(qdc): split QDC into a PR smoke tier and a nightly full tier

  #1601·Mengsheng Wu (mengshengwu) opened Oct 26, 2026Contributor·5 tasks done·

  ·
  1

  

• perf(sdk): row-split llama_cpp HTP devices across both NPUs on IQ9

  #1600·Mengsheng Wu (mengshengwu) opened Oct 26, 2026Contributor·4 tasks done·

  ·
  1

  

• chore(sdk): bump llama.cpp submodule to d854d98ba

  #1599·Mengsheng Wu (mengshengwu) opened Oct 26, 2026Contributor·3 of 4 tasks·

  ·
  1

  

• fix(sdk): synchronize Qwen3.5 MTP drafter state

  #1595·Jaehyup (Jay) Shin (shin0343) opened Oct 23, 2026Contributor·7 of 8 tasks·

  ·
  1

  1

• feat(sdk): ship Adreno OpenCL bin kernel library on Windows

  #1594·zdc (Hahahaooovvv) opened Oct 23, 2026Contributor·4 of 8 tasks·

  ·

• fix(sdk): use Q8_0 KV and dual-NPU HTP layout for llama_cpp NPU

  #1592·Mengsheng Wu (mengshengwu) opened Oct 23, 2026Contributor·1 of 4 tasks·

  ·
  

• feat(sdk): add streaming Phonon QNN TTS support

  #1581·Shivay Lamba (shivaylamba) opened Oct 22, 2026·9 of 14 tasks·

  ·

• ci(release): publish standalone llama_cpp plugin zip for windows-arm64

  #1576·Perry Cheng (zhycheng614) opened Oct 20, 2026Contributor·2 tasks·

  ·
  1

• docs(release): sign only the release-built HTP to-sign zip

  #1563·Perry Cheng (zhycheng614) opened Oct 19, 2026Contributor·

  ·
  1

• fix(sdk): remove genie_config.json dependency in qairt plugin

  #1522·Perry Cheng (zhycheng614) opened Oct 8, 2026Contributor·

  ·

• feat(sdk): accept prompt_utf8 in geniex_llm_forward_logits

  #1519·Shishir Shrivastava (Shishir2405) opened Oct 7, 2026·2 of 5 tasks·

  ·
  1

  1

• test(sdk): add Windows HTP preflight

  #1511·Khoi VN (vnk8071) opened Oct 5, 2026Member·2 of 3 tasks·

  ·
  1

• fix(cli): trim old run turns when context fills

  #1506·MAN$I VERMA 🦀 (mansiverma897993) opened Oct 3, 2026Contributor·4 of 5 tasks·

  ·
  1

• fix(sdk): prefetch the right weight chunk in batched F16 HMX matmul

  #1505·Tadayuki Okada (TadayukiOkada) opened Oct 2, 2026·2 of 3 tasks·

  ·

• fix(sdk): accept a pasted model URL in geniex_model_get_paths

  #1502·Alexandre Kohler (kwy404) opened Oct 1, 2026·4 tasks done·

  ·

• feat(sdk): run llama_cpp HTP on Hexagon SDK 5.5.x functional-safety firmware

  #1499·Tadayuki Okada (TadayukiOkada) opened Sep 30, 2026·12 of 13 tasks·

  ·

• fix(sdk)!: support multimodal KV prefix reuse

  #1495·MAN$I VERMA 🦀 (mansiverma897993) opened Sep 29, 2026Contributor·

  ·
  1

• feat(sdk): enable experimental Windows x64 CUDA builds

  Stale
  #1474·Sam (samshanmukh) opened Sep 21, 2026·7 of 9 tasks·

  ·
  1

• fix(qairt): use host workers for decode KV overlap

  Stale
  #1440·MAN$I VERMA 🦀 (mansiverma897993) opened Sep 8, 2026Contributor·

  ·
  2

• feat(server): add transactional managed cache lineage

  Stale
  #1414·Sam Kwak (samkwak188) opened Aug 31, 2026·11 of 13 tasks·

  ·
  1

  1

• fix(release): sign Windows ARM64 binaries for Smart App Control

  Stale
  #1410·MAN$I VERMA 🦀 (mansiverma897993) opened Aug 29, 2026Contributor·4 of 5 tasks·

  ·
  1

  1

• Fix Gemma 4 required tool selection and native call normalization

  Stale
  #1337·GlennLin opened Aug 13, 2026·

  ·
  1

• fix(sdk): harden model lifecycle and Xray validation

  Stale
  #1323·Kristoffer Södersten (kristoffersodersten) opened Aug 11, 2026·

  ·
  1