yuan2

Yuan2

In this directory, you will find examples on how you could apply IPEX-LLM INT4 optimizations on Yuan2 models on Intel GPUs. For illustration purposes, we utilize the IEITYuan/Yuan2-2B-hf as a reference Yuan2 model.

0. Requirements

To run these examples with IPEX-LLM, we have some recommended requirements for your machine, please refer to here for more information.

In addition, you need to modify some files in Yuan2-2B-hf folder, since Flash attention dependency is for CUDA usage and currently cannot be installed on Intel CPUs. To manually turn it off, please refer to this issue. We also provide two modified files(config.json and yuan_hf_model.py), which can be used to replace the original content in config.json and yuan_hf_model.py.

Example: Predict Tokens using generate() API

In the example generate.py, we show a basic use case for an Yuan2 model to predict the next N tokens using generate() API, with IPEX-LLM INT4 optimizations on Intel GPUs.

1. Install

1.1 Installation on Linux

We suggest using conda to manage environment:

conda create -n llm python=3.9
conda activate llm
# below command will install intel_extension_for_pytorch==2.1.10+xpu as default
pip install --pre --upgrade ipex-llm[all] # install the latest ipex-llm nightly build with 'all' option
pip install einops # additional package required for Yuan2 to conduct generation
pip install pandas # additional package required for Yuan2 to conduct generation

1.2 Installation on Windows

We suggest using conda to manage environment:

conda create -n llm python=3.9 libuv
conda activate llm
# below command will install intel_extension_for_pytorch==2.1.10+xpu as default
pip install --pre --upgrade ipex-llm[xpu] -f https://developer.intel.com/ipex-whl-stable-xpu
pip install einops # additional package required for Yuan2 to conduct generation

2. Configures OneAPI environment variables

2.1 Configurations for Linux

source /opt/intel/oneapi/setvars.sh

2.2 Configurations for Windows

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

Note: Please make sure you are using CMD (Anaconda Prompt if using conda) to run the command as PowerShell is not supported.

3. Runtime Configurations

For optimal performance, it is recommended to set several environment variables. Please check out the suggestions based on your device.

3.1 Configurations for Linux

For Intel Arc™ A-Series Graphics and Intel Data Center GPU Flex Series

export USE_XETLA=OFF
export SYCL_PI_LEVEL_ZERO_USE_IMMEDIATE_COMMANDLISTS=1

For Intel Data Center GPU Max Series

export LD_PRELOAD=${LD_PRELOAD}:${CONDA_PREFIX}/lib/libtcmalloc.so
export SYCL_PI_LEVEL_ZERO_USE_IMMEDIATE_COMMANDLISTS=1
export ENABLE_SDP_FUSION=1

Note: Please note that libtcmalloc.so can be installed by conda install -c conda-forge -y gperftools=2.10.

#### 3.2 Configurations for Windows

For Intel iGPU

set SYCL_CACHE_PERSISTENT=1
set BIGDL_LLM_XMX_DISABLED=1

For Intel Arc™ A300-Series or Pro A60

set SYCL_CACHE_PERSISTENT=1

For other Intel dGPU Series

There is no need to set further environment variables.

Note: For the first time that each model runs on Intel iGPU/Intel Arc™ A300-Series or Pro A60, it may take several minutes to compile.

4. Running examples

python ./generate.py

In the example, several arguments can be passed to satisfy your requirements:

• --repo-id-or-model-path REPO_ID_OR_MODEL_PATH: argument defining the huggingface repo id for the Yuan2 model to be downloaded, or the path to the huggingface checkpoint folder. It is default to be 'IEITYuan/Yuan2-2B-hf'.
• --prompt PROMPT: argument defining the prompt to be infered (with integrated prompt format for chat). It is default to be 'What is AI?'.
• --n-predict N_PREDICT: argument defining the max number of tokens to predict. It is default to be 100.

Sample Output

IEITYuan/Yuan2-2B-hf

Inference time: xxxx seconds
-------------------- Output --------------------

What is AI?
AI is a field of technology and technologies that is used to analyze and improve human behavior such as language processing, machine learning and artificial intelligence (AI).<eod>