A "SYCL error" happened when load Phi3 model to Intel GPU

[soulyet]

We follow the guideline to setup IPEX-LLM[CPP] and also Intel OneAPI. but when we load Phi3 model to Intel GPU, it always have "SYCL error" happened. and we also use main.exe to load model manually, same error.
I attached ollama and llama.cpp log files:
llama.cpp_log.txt
ollama_log.txt

I'm not sure is there is any environment issue for my setup.

[rnwang04]

Hi @soulyet ,
Q4_K has some error at your current machine, we may try to fix it later.
But for now, I think you can try Q4_0 first, which should work.

[soulyet]

Hi @soulyet , Q4_K has some error at your current machine, we may try to fix it later. But for now, I think you can try Q4_0 first, which should work.

Thanks for reply.
Now I use Q4_0 which download from https://huggingface.co/SanctumAI/Phi-3-mini-4k-instruct-GGUF/tree/main for try, but met another error:
GGML_ASSERT: C:\Users\Administrator\actions-runner\cpp-release_work\llm.cpp\llm.cpp\ollama-internal\llm\llama.cpp\ggml-backend.c:100: base != NULL && "backend buffer base cannot be NULL"

log detail:
llama_log.txt