experiments/lm-harness-evaluation.md

LM Harness Evaluation

The evaluation harness from EleutherAI is integrated a submodule. We use a fork on HF's Github. To initialize the submodule, run:

git submodule init
git submodule update

Make sure you have the requirements in lm-evaluation-harness:

cd lm-evaluation-harness
pip install -r requirements.txt

To launch an evaluation, run:

python lm-evaluation-harness/main.py \
    --model gpt2 \
    --model_args pretrained=gpt2-xl \
    --tasks cola,mrpc,rte,qnli,qqp,sst,boolq,cb,copa,multirc,record,wic,wsc,coqa,drop,lambada,lambada_cloze,piqa,pubmedqa,sciq \
    --provide_description \ # Whether to provide the task description
    --num_fewshot 3 \ # Number of priming pairs
    --batch_size 2 \
    --output_path eval-gpt2-xl

Please note:

• As of now, only single GPU is supported in lm-evaluation-harness.