Intel/falcon-7b-sq-int8-inc

Model Details

This model is an int8 model quantized from tiiuae/falcon-7b using SmoothQuant.

Env setup

Environment Setup

Inferece

Use IPEX 2.2

git clone https://github.com/intel/intel-extension-for-pytorch.git
cd intel-extension-for-pytorch/examples/cpu/inference/python/llm
git checkout release/2.2
python run.py  --benchmark -m tiiuae/falcon-7b --ipex-smooth-quant --qconfig-summary-file <path to Intel/falcon-7b-sq-int8-inc best_configure.json"> --output-dir "saved_results"

Evaluate

Evaluate the model

git clone https://github.com/intel/intel-extension-for-pytorch.git
cd intel-extension-for-pytorch/examples/cpu/inference/python/llm/single_instance
git checkout release/2.2
python run_accuracy.py -m tiiuae/falcon-7b --quantized-model-path <path to Intel/falcon-7b-sq-int8-inc best_configure.json"> --dtype int8  --tasks lambada_openai

Results

Metric	fp32	int8 sq
Avg.	0.6982	0.6992
lambada_openai	0.7467	0.7648
hellaswag	0.5778	0.5659
winogrande	0.6732	0.6717
piqa	0.7949	0.7943

Ethical Considerations and Limitations

The model can produce factually incorrect output, and should not be relied on to produce factually accurate information. Because of the limitations of the pretrained model and the finetuning datasets, it is possible that this model could generate lewd, biased or otherwise offensive outputs.

Therefore, before deploying any applications of the model, developers should perform safety testing.

Caveats and Recommendations

Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model.

Here are a couple of useful links to learn more about Intel's AI software:

• Intel® Neural Compressor
• Intel® Extension for PyTorch