the reference results are blank with deepseek model and our generate example code

[K-Alex13]

Following the guild of https://github.com/intel-analytics/ipex-llm/tree/main/python/llm/example/GPU/HuggingFace/LLM/deepseek
and change the model to deepseek-ai/deepseek-coder-1.3b-instruct
following is the results

please help me

[K-Alex13]

even try to use original 6.7B model, a new problem come up

[Oscilloscope98]

Hi @K-Alex13,

We are reproducing this issue, and will let you know for any updates :)

[Oscilloscope98]

Hi @K-Alex13,

We have reproduced this issue and currently fixing it. We will update here for any progress :)

[Oscilloscope98]

Hi @K-Alex13,

We have fixed this blank output issue for deepseek-coder models, you could upgrade to ipex-llm>=2.2.0b20250115 and have a try again with the updated deepseek GPU example.

For your second issue regarding deepseek-ai/deepseek-coder-6.7b-instruct, this is because your memory is insufficient to load the original 6.7b model for further IPEX-LLM low-bit optimizations. You can use a machine with sufficient memory to save the IPEX-LLM low-bit model, then load the low-bit model onto the current machine for inference. You could refer to our Save-Load example for more information.

If no machine with larger memory is available, you could also try increasing your virtual memory to load the original mode.

Please let us know for any further problems :)