Issues

• v0.10.0: LFM3-Omni-2B reports a good load, then crashes on any prompt

  qualcomm/GenieX#1656 · Filter by author jdtimothy opened on Nov 5, 2026

• v0.10.0: vision models crash on the first image on Adreno 840 GPU (Galaxy S26 Ultra)

  qualcomm/GenieX#1655 · Filter by author jdtimothy opened on Nov 5, 2026

• [llama.cpp / NPU] Windows on Snapdragon X (X1-26-100): hard system reset (hardware watchdog, Kernel-Power 142) during prompt prefill; QAIRT unaffected

  qualcomm/GenieX#1652 · Filter by author renaissancexd opened on Nov 3, 2026
  3

• LFM3 runs slower on the NPU after a QAIRT model has run, until reboot

  qualcomm/GenieX#1650 · Filter by author sanoop-sadheerthan opened on Nov 3, 2026

• [llama.cpp / NPU] serve hangs forever in ggml_hexagon_session::flush_pending (dspqueue_read)

  qualcomm/GenieX#1649 · Filter by author sanoop-sadheerthan opened on Nov 3, 2026

• v0.10.0: LFM2-VL / LFM2.5-VL generates false caption for vision encoder on NPU

  qualcomm/GenieX#1648 · Filter by author sanoop-sadheerthan opened on Nov 3, 2026

• Make the context window for QAIRT bigger

  qualcomm/GenieX#1647 · Filter by author josergc opened on Nov 3, 2026

• serve: non-streaming chat completions always include empty function_call, audio, refusal — clients that honour legacy function_call break

  qualcomm/GenieX#1646 · Filter by author john-owens opened on Oct 31, 2026
  1

• v0.10.0 regression on Windows-on-ARM: Qwen3-VL vision on HTP0 silently returns garbage (the #1476 fix from v0.9.3 does not hold)

  qualcomm/GenieX#1633 · Filter by author maxl668 opened on Oct 29, 2026
  2

  

• ggml-hex reports an opaque 'failed to open session ... error 0x72' when the CDSP is offline

  qualcomm/GenieX#1607 · Filter by author sirredbeard opened on Oct 27, 2026
  

• v0.9.3: first chipset lookup blocks for over a minute on a device without internet access (platform.json retries)

  qualcomm/GenieX#1605 · Filter by author kvapr opened on Oct 26, 2026
  

• Nemotron 3 Nano 30B-A3B (GGUF Q4_0) produces gibberish on both NPU and CPU, while upstream llama.cpp is fine

  qualcomm/GenieX#1604 · Filter by author sjoffraud opened on Oct 26, 2026
  1

  

• Qwen3-VL-8B-Instruct fails to execute graph with Error 1007 (DSP CONN Reset / SSR) on Snapdragon X Plus Surface Pro 11

  qualcomm/GenieX#1596 · Filter by author FuzzyIkigai opened on Oct 24, 2026
  

• llama_cpp plugin on Windows: CPU threadpool spin-waits (ggml poll 50) through every token while the model runs on the NPU

  qualcomm/GenieX#1590 · Filter by author Jcollier0120 opened on Oct 22, 2026
  

• Consecutive request failures in Qwen 3.5

  qualcomm/GenieX#1558 · Filter by author zhuxiaodongzxd opened on Oct 18, 2026
  1

  

• llama_cpp plugin (Windows, NPU + OpenCL vision): ~1.4 GB more memory than needed for Qwen3-VL-2B

  qualcomm/GenieX#1557 · Filter by author Jcollier0120 opened on Oct 17, 2026
  

• [Feature Request] qairt VLMs: load the embedding table quantized and memory-mapped, as text LLMs and Gemma4 already do

  qualcomm/GenieX#1547 · Filter by author Jcollier0120 opened on Oct 15, 2026
  2

  

• How could we sign libggml-htp.cat with Microsoft Windows Hardware Compatibility Publisher

  qualcomm/GenieX#1542 · Filter by author zhang261007 opened on Oct 15, 2026
  1

• bug(bench): Gemma-4-E4B HTP benchmark fails FastRPC mapping with default settings on SM8650

  qualcomm/GenieX#1531 · Filter by author mansiverma897993 opened on Oct 13, 2026
  5

  

• Ternary (1.58-bit) GGUF from Hugging Face: which files the pinned llama.cpp refuses, and an offer to help

  qualcomm/GenieX#1515 · Filter by author dmitrii-f-t27 opened on Oct 6, 2026
  2

  

• QAIRT Qwen3-4B-Instruct-2507: decode ~30x slower when the context binary holds 10 graphs per part; same graphs relinked as ≤8 per part run at full speed (Snapdragon X X1-26-100)

  qualcomm/GenieX#1509 · Filter by author ylu861 opened on Oct 4, 2026
  4

• [Bug] ggml-hex (HTP): Silent corruption + dspqueue Error 0xc timeouts on Hexagon v73 (SM7635)

  qualcomm/GenieX#1508 · Filter by author algo1127 opened on Oct 3, 2026
  3

  

• v0.8.0 Android/SM8650: HTP session fails to open with error 0x80000406

  qualcomm/GenieX#1507 · Filter by author crab3e opened on Oct 3, 2026
  3

  

• [Feature Request]: received error while streaming: SDKError(Input prompt too long)

  qualcomm/GenieX#1498 · Filter by author josergc opened on Sep 30, 2026
  2

  

• Phi-3.5-Mini-Instruct doesn't load in Elite X (12 cores) X1E80100 Hexagon NPU

  qualcomm/GenieX#1494 · Filter by author josergc opened on Sep 28, 2026
  1