[SYCL]Sysman free-memory query may be unavailable

[DDXDB]

Name and Version

version: 0.3.0-dev (build 10759, commit b81c99b)
built with Clang 20.1.8 for Windows x86_64

Operating systems

Windows

Which llama.cpp modules do you know to be affected?

llama-server

Command line

llama\llama-server.exe -m llama\Huihui-Qwen3.8-27B-abliterated-Q6_K.gguf --host 0.0.0.0 --n-gpu-layers -1  --reasoning off  --reasoning-effort  minimal --reasoning-budget 0 --temp 0.6 --top-k 20 --top-p 0.95 -c 40960 -n 32768  --min-p 0.0   --alias ALIAS --timeout 99999999 --cpu-moe --log-file log.txt  --spec-type draft-mtp,ngram-map-k4v,ngram-mod -sm layer --load-mode auto -ctk q8_0 -ctv q8_0 --jinja

Problem description & steps to reproduce

In the console, the message "Sysman free-memory query may be unavailable" keeps recurring（Not appearing in the log）.
And after some time, error

Warning: zesInit failed [ggml_check_sycl] with code 2013265921. Sysman free-memory query may be unavailable.
Warning: zesInit failed [ggml_check_sycl] with code 2013265921. Sysman free-memory query may be unavailable.
Warning: zesInit failed [ggml_check_sycl] with code 2013265921. Sysman free-memory query may be unavailable.
Warning: zesInit failed [ggml_check_sycl] with code 2013265921. Sysman free-memory query may be unavailable.
Warning: zesInit failed [ggml_check_sycl] with code 2013265921. Sysman free-memory query may be unavailable.
Warning: zesInit failed [ggml_check_sycl] with code 2013265921. Sysman free-memory query may be unavailable.
Warning: zesInit failed [ggml_check_sycl] with code 2013265921. Sysman free-memory query may be unavailable.
3.50.265.387 W ggml_sycl_flash_attn_ext_onednn: oneDNN SDPA failed (level_zero backend failed with error: 20 (UR_RESULT_ERROR_DEVICE_LOST)); falling back to TILE kernel
Warning: zesInit failed [ggml_check_sycl] with code 2013265921. Sysman free-memory query may be unavailable.
3.50.267.586 E srv  update_slots: decode() failed: level_zero backend failed with error: 2147483646 (UR_RESULT_ERROR_UNKNOWN)
3.50.267.590 E srv    send_error: task id = 10, error: decode() failed: level_zero backend failed with error: 2147483646 (UR_RESULT_ERROR_UNKNOWN)
3.50.267.597 I slot      release: id  3 | task 10 | stop processing: n_tokens = 958, truncated = 0
3.50.267.609 W srv          stop: cancel task, id_task = 10

First Bad Commit

a32af33

Relevant log output

Logs

[log.txt](https://github.com/user-attachments/files/31734482/log.txt)

[arthw]

@DDXDB
#28227 fix it.
Please try it!

Thank you!

[DDXDB]

@DDXDB #28227 fix it. Please try it!

Thank you!

I tried, and the result was:

Active code page: 65001
:: initializing oneAPI environment...
   Initializing Visual Studio command-line environment...
   Visual Studio version 18.9.2 environment configured.
   "C:\Program Files\Microsoft Visual Studio\18\Community\"
   Visual Studio command-line environment initialized for: 'x64'
:  compiler -- latest
:  dev-utilities -- latest
:  dpl -- latest
:  mkl -- latest
:  ocloc -- latest
:  pti -- latest
:  tbb -- latest
:  tcm -- latest
:  umf -- latest
:: oneAPI environment initialized ::
0.00.168.459 I cmn  common_param: common_params_print_info: verbosity = 3 (adjust with the `-lv N` CLI arg)
0.00.168.643 W srv  llama_server: -----------------
0.00.168.644 W srv  llama_server: CORS is set to allow all origins ('*') and no API key is set
0.00.168.645 W srv  llama_server: this can be a security risk (cross-origin attacks)
0.00.168.645 W srv  llama_server: more info: https://github.com/ggml-org/llama.cpp/pull/25655
0.00.168.645 W srv  llama_server: -----------------
0.00.175.406 I srv    load_model: loading model 'llama\Huihui-Qwen3.8-27B-abliterated-Q6_K.gguf'
F:\llama.cpp\ggml\src\ggml.c:3953: GGML_ASSERT(a->ne[0] == b->ne[0]) failed

Use

call "C:\Program Files (x86)\Intel\oneAPI\setvars.bat" intel64

llama\llama-server.exe -m llama\Huihui-Qwen3.8-27B-abliterated-Q6_K.gguf --host 0.0.0.0 --n-gpu-layers -1  --reasoning off  --reasoning-effort  minimal --reasoning-budget 0 --temp 0.6 --top-k 20 --top-p 0.95 -c 40960 -n 32768  --min-p 0.0   --alias ALIAS --timeout 99999999 --cpu-moe --log-file log.txt  --spec-type draft-mtp,ngram-map-k4v,ngram-mod -sm layer --load-mode auto -ctk q8_0 -ctv q8_0 --jinja