logical-reasoning / scripts /eval-mgtv-llama3_8b.sh
dh-mc's picture
ready for llama3 r4
fcc0071
#!/bin/sh
BASEDIR=$(dirname "$0")
cd $BASEDIR/..
echo Current Directory:
pwd
BASEDIR=`pwd`
nvidia-smi
uname -a
cat /etc/os-release
lscpu
grep MemTotal /proc/meminfo
#pip install -r requirements.txt
#cd ../LLaMA-Factory && pip install -e .[torch,bitsandbytes]
export LOGICAL_REASONING_DATA_PATH=datasets/mgtv
export RESIZE_TOKEN_EMBEDDINGS=true
export START_EPOCH=0
export USING_LLAMA_FACTORY=true
export MODEL_NAME=shenzhi-wang/Llama3-8B-Chinese-Chat
# export MODEL_NAME=hfl/llama-3-chinese-8b-instruct-v3
export MODEL_PREFIX=llama3-8b_lora_sft_bf16
export LOGICAL_REASONING_RESULTS_PATH=results/$MODEL_PREFIX-p1_r4.csv
export ADAPTER_PATH_BASE=llama-factory/saves/llama3-8b/lora/sft_bf16_p1_full_r4
export USING_P1_PROMPT_TEMPLATE=true
echo "Eval $MODEL_NAME with $ADAPTER_PATH_BASE"
python llm_toolkit/eval_logical_reasoning_all_epochs.py
# export LOGICAL_REASONING_RESULTS_PATH=results/$MODEL_PREFIX-p2_r4.csv
# export ADAPTER_PATH_BASE=llama-factory/saves/llama3-8b/lora/sft_bf16_p2_full_r4
# export USING_P1_PROMPT_TEMPLATE=false
# echo "Eval $MODEL_NAME with $ADAPTER_PATH_BASE"
# python llm_toolkit/eval_logical_reasoning_all_epochs.py