flan-t5

Flan-t5

In this directory, you will find examples on how you could apply IPEX-LLM INT4 optimizations on Flan-t5 models. For illustration purposes, we utilize the google/flan-t5-xxl as a reference Flan-t5 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.

Example: Predict Tokens using generate() API

In the example generate.py, we show a basic use case for a Flan-t5 model to predict the next N tokens using generate() API, with IPEX-LLM INT4 optimizations.

1. Install

We suggest using conda to manage the Python environment. For more information about conda installation, please refer to here.

After installing conda, create a Python environment for IPEX-LLM:

On Linux:

conda create -n llm python=3.11 # recommend to use Python 3.11
conda activate llm

# install the latest ipex-llm nightly build with 'all' option
pip install --pre --upgrade ipex-llm[all] --extra-index-url https://download.pytorch.org/whl/cpu

On Windows:

conda create -n llm python=3.11
conda activate llm

pip install --pre --upgrade ipex-llm[all]

2. Run

After setting up the Python environment, you could run the example by following steps.

Note: When loading the model in 4-bit, IPEX-LLM converts linear layers in the model into INT4 format. In theory, a XB model saved in 16-bit will requires approximately 2X GB of memory for loading, and ~0.5X GB memory for further inference.

Please select the appropriate size of the Flan-t5 model based on the capabilities of your machine.

2.1 Client

On client Windows machines, it is recommended to run directly with full utilization of all cores:

python ./generate.py --prompt 'Translate to German: My name is Arthur'

More information about arguments can be found in Arguments Info section. The expected output can be found in Sample Output section.

2.2 Server

For optimal performance on server, it is recommended to set several environment variables (refer to here for more information), and run the example with all the physical cores of a single socket.

E.g. on Linux,

# set IPEX-LLM env variables
source ipex-llm-init

# e.g. for a server with 48 cores per socket
export OMP_NUM_THREADS=48
numactl -C 0-47 -m 0 python ./generate.py --prompt 'Translate to German: My name is Arthur'

More information about arguments can be found in Arguments Info section. The expected output can be found in Sample Output section.

2.3 Arguments Info

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 Flan-t5 model (e.g. google/flan-t5-xxl) to be downloaded, or the path to the huggingface checkpoint folder. It is default to be 'google/flan-t5-xxl'.
• --prompt PROMPT: argument defining the prompt to be infered (with integrated prompt format for chat). It is default to be 'Translate to German: My name is Arthur'.
• --n-predict N_PREDICT: argument defining the max number of tokens to predict. It is default to be 32.

2.4 Sample Output

google/flan-t5-xxl

Inference time: xxxx s
-------------------- Prompt --------------------
<|User|>:Translate to German: My name is Arthur
-------------------- Output --------------------
Ich bin Arthur.