eval_layer_llama.sh

export HTTP_PROXY=http://sys-proxy-rd-relay.byted.org:8118
export http_proxy=http://sys-proxy-rd-relay.byted.org:8118
export https_proxy=http://sys-proxy-rd-relay.byted.org:8118
export no_proxy="$no_proxy,.byteintl.net"
export HF_ENDPOINT=https://hf-mirror.com


cd quantization_metric/
cuda_id=0
model_id="/mnt/bn/life-mllm/users/cxr/quantization/models/meta-llama/Llama-3.1-8B"
model_name=$(basename "$model_id")

model=/mnt/bn/life-mllm/users/cxr/quantization/models/${model_name}-quantization-layer
# output_dir=Alpha_values_mlp
# bit_layers_dir=/mnt/bn/life-mllm/users/cxr/quantization/quantization_metric/bit_layers
# result_dir=/mnt/bn/life-mllm/users/cxr/quantization/lm-evaluation-harness/results



start=$(date +%s.%N)
# rm -rf $model

modes=("self_attn" "mlp")
for mode in ${modes[@]}; do
    for idx in {-1..31}; do
        echo $mode $idx
        cd ../quantization_metric
        python -u main.py --bit_layer_idx $idx  --save_dir ${model} --mode $mode --model_id $model_id --cuda_id $cuda_id
        cd ../lm-evaluation-harness
        bash run_scripts/eval_base_qwen2_5_7b.sh ${model} ${mode}_${idx} $cuda_id
        rm -rf ${model}
        end=$(date +%s.%N)
        runtime=$(awk "BEGIN {print $end - $start}")
        echo "Execution time: $runtime seconds"

    done
done