[SYCL] Bug: check_allow_gpu_index error: device_index:0 is out of range: [0--2]

[nickp27]

What happened?

SYCL bug when trying to use it as an RPC backend.

Seems to be the same bug as present in whisper.cpp - ggerganov/whisper.cpp#2033

Will keep digging around to find the cause but posting in the meantime.

Name and Version

[b3071] (https://github.com/ggerganov/llama.cpp/releases/tag/b3071)

What operating system are you seeing the problem on?

Linux

Relevant log output

create_backend: using SYCL backend
[SYCL] call ggml_init_sycl
ggml_init_sycl: GGML_SYCL_DEBUG: 0
ggml_init_sycl: GGML_SYCL_F16: yes
found 4 SYCL devices:
|  |                   |                                       |       |Max    |        |Max  |Global |                     |
|  |                   |                                       |       |compute|Max work|sub  |mem    |                     |
|ID|        Device Type|                                   Name|Version|units  |group   |group|size   |       Driver version|
|--|-------------------|---------------------------------------|-------|-------|--------|-----|-------|---------------------|
| 0| [level_zero:gpu:0]|                  Intel HD Graphics 630|    1.3|     24|     256|   32| 62191M|            1.3.27642|
| 1|     [opencl:gpu:0]|                  Intel HD Graphics 630|    3.0|     24|     256|   32| 62191M|       23.43.27642.40|
| 2|     [opencl:cpu:0]|       Intel Core i5-7500 CPU @ 3.40GHz|    3.0|      4|    8192|   64| 66662M|2024.17.3.0.08_160000|
| 3|     [opencl:acc:0]|            Intel FPGA Emulation Device|    1.2|      4|67108864|   64| 66662M|2024.17.3.0.08_160000|
check_allow_gpu_index error: device_index:0 is out of range: [0--2]
check_allow_gpu_index error: device_index:0 is out of range: [0--2]
8:02PM INF Starting llama-cpp-rpc-server on '127.0.0.1:33521'

[rgerganov]

As @slaren pointed out in #7808 the SYCL backend is using init_tensor do set tensor extra properties and this is not currently supported by the RPC backend.

[github-actions]

This issue was closed because it has been inactive for 14 days since being marked as stale.