scripts/eval-shots.sh

#!/bin/sh

BASEDIR=$(dirname "$0")
cd $BASEDIR/..
echo Current Directory:
pwd

export LOGICAL_REASONING_DATA_PATH=datasets/mgtv
export RESIZE_TOKEN_EMBEDDINGS=true
# export USING_LLAMA_FACTORY=true
export USING_P1_PROMPT_TEMPLATE=false
export LOAD_IN_4BIT=false

export ORG_NAME=$1
export MODEL=$2
export MODEL_NAME=$ORG_NAME/$MODEL

export LOGICAL_REASONING_RESULTS_PATH=data/${MODEL}_results.csv

if [ "$MODEL" == "Qwen2.5-3B-Instruct" ];
then
    echo "Skipping Qwen2.5-3B-Instruct"
    exit 0
fi

echo Evaluating $MODEL_NAME with few-shot learning
python llm_toolkit/eval_shots.py

if [ "$MODEL" == "internlm2_5-7b-chat-1m" ];
then
    export LOAD_IN_4BIT=true
    export START_NUM_SHOTS=5
    $BASEDIR/scripts/eval-shots.sh internlm internlm2_5-20b-chat
fi