logical-reasoning / scripts /tune-mgtv-gemma2_9b.sh
inflaton's picture
attempted gemma2
bc0d2ff
#!/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] && cd $BASEDIR
#pip install -U transformers
#pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu121
export LOGICAL_REASONING_DATA_PATH=datasets/mgtv
export MODEL_NAME=shenzhi-wang/Gemma-2-9B-Chinese-Chat
export MODEL_PREFIX=gemma2-9b_lora_sft_bf16
export CONFIG_FILE=config/$MODEL_PREFIX-p1.yaml
echo "Tuning with $CONFIG_FILE"
$BASEDIR/scripts/tune-lf.sh $CONFIG_FILE
export LOGICAL_REASONING_RESULTS_PATH=results/$MODEL_PREFIX-p1.csv
export ADAPTER_PATH_BASE=llama-factory/saves/qwen2-72b/lora/sft_4bit_p1_full
echo "Eval $MODEL_NAME with $ADAPTER_PATH_BASE"
python llm_toolkit/eval_logical_reasoning_all_epochs.py
export CONFIG_FILE=config/$MODEL_PREFIX-p2.yaml
echo "Tuning with $CONFIG_FILE"
$BASEDIR/scripts/tune-lf.sh $CONFIG_FILE
export LOGICAL_REASONING_RESULTS_PATH=results/$MODEL_PREFIX-p2.csv
export ADAPTER_PATH_BASE=llama-factory/saves/qwen2-72b/lora/sft_4bit_p2_full
echo "Eval $MODEL_NAME with $ADAPTER_PATH_BASE"
python llm_toolkit/eval_logical_reasoning_all_epochs.py