Ruonan Wang

Hi @doublefx , I want to confirm that, are you already changing the NPU driver to [`32.0.100.3104 `](https://www.intel.com/content/www/us/en/download/794734/838895/intel-npu-driver-windows.html), and still got above errors ?

Hi @doublefx , just share some personal experience here. First about the installation part, I just ```bash conda create -n llm-npu python=3.11 conda activate llm-npu pip install --pre --upgrade ipex-llm[npu]...

Just two reminders, maybe you can check if you have enough disk space `C:\` and check if `IPEX_LLM_NPU_MTL=1` really works, for example, if you are running with Miniforge Prompt ....

Hi @jaredpilcher , if you want to reduce num_parallel to save kv cache memory, you could `set OLLAMA_NUM_PARALLEL=1` in your cmd.

For the dark screen issue, you may need update mesa related package: `apt install libegl-mesa0 libglapi-mesa libgl1-mesa-dri libglu1-mesa libglx-mesa0 mesa-vulkan-drivers`

Hi @markussiebert , it looks like not an expected result. On our A770 machine, with https://github.com/ipex-llm/ipex-llm/releases/download/v2.3.0-nightly/llama-cpp-ipex-llm-2.3.0b20250612-ubuntu-core.tgz, for DeepSeek-R1-0528-Qwen3-8B-UD-Q8_K_XL.gguf I can get about 30 tokens/second for decode stage. My test command...

Hi @markussiebert , ### performance on my machine My test machine is i9-13900K + A770, and I tested with this portable zip https://github.com/ipex-llm/ipex-llm/releases/download/v2.3.0-nightly/llama-cpp-ipex-llm-2.3.0b20250612-ubuntu-core.tgz . On this machine, I got about...

Hi @Mailootje , could you please provide us with more machine info ? You could check it with [env-check.sh](https://github.com/intel/ipex-llm/blob/main/python/llm/scripts/env-check.sh) in https://github.com/intel/ipex-llm/tree/main/python/llm/scripts and show us the output.

Hi @Teejer , could you please upgrade ollama version first by `pip install ipex-llm[cpp]==2.3.0b20250529` and try it again ?

Hi @thehoff , `2.3.0rc1` is an older version, please try `pip install ipex-llm[cpp]==2.3.0b20250530` . The information here is somewhat misleading, and we will consider optimizing this part. In fact, if...