This repository contains the code for the paper LogisticsLLM: Network Freight Price Prediction Based on Large Language Models. This paper introduces Large Language Models (LLMs) to predict network freight prices by using prior knowledge inherent in the LLMs.

python data_process.py --mode=SI --dataset=MC --order_strategy=InitOrd --sequence=NamedFeatSeq --split=82

python data_process.py --mode=SI --dataset=HE --order_strategy=SeqeOrd --sequence=NamedFeatSeq --split=82

CUDA_VISIBLE_DEVICES=0 llamafactory-cli train config/MC_Scale100_Split82_InitOrd_ValueOnlySeq_Cot0_GLM4-9b-chat_Epochs5_En.yaml

python generate_cross_parameter_yaml.py

python generate_bash_script.py

nohup ./scripts/train.sh > logs/output.log 2>&1 &

pkill -f train.sh

python llm_logistics.py --dataset=MC --model=GLM4-9b-chat --epoch=5 --mode=test --loss_fun=cross --split=82 --order_strategy=InitOrd --sequence=NamedFeatSeq --cot=0 --scale=100

python llm_logistics.py --dataset=HE --model_name=GLM4-9b-chat --epoch=15 --loss_fun=cross --mode=test --split=82 --order_strategy=InitOrd --sequence=ValueOnly --cot=0 --scale=100

nohup ./scripts/predict.sh > logs/output.log 2>&1 &

• BA (batch)
• SI (single)

• MC (MathCup)
• HE (HackerEarth)

• InitOrd (init)
• SeqeOrd (feature)
• DistOrd (distance)

• NamedFeatSeq (FeatureName: FeatureValue, FeatureName: FeatureValue)
• ValueOnlySeq (FeatureValue FeatureValue)
• ListTempSeq (FeatureName: FeatureValue\n FeatureName: FeatureValue)
• JsonSeq ("FeatureName": "FeatureValue", "FeatureName": "FeatureValue")

• GLM4-9b-chat
• Qwen2-7B-Instruct
• Llama-3.1-8B-Instruct

• Epochs5
• Epochs10

• checkpoints: checkpoints/HE_Split82_DistOrd_ListTempSeq_Cot0_GLM4-9b-chat_Epochs5_En
• train_file: HE_Split82_DistOrd_ListTempSeq_Cot0_En_train.json
• test_file: HE_Split82_DistOrd_ListTempSeq_Cot0_En_test.json
• result_file: MC_Split82_DistOrd_NamedFeatSeq_Cot0_GLM4-9b-chat_Epochs3_En_20240820.csv

The repo structure and module functions are as follows:

project_name/
│
├── config/                                                 // training paramer file
│   ├── HE_Scale100_Split82_DistOrd_JsonSeq_Cot0_llama-3-8b-instruct_Epochs15_En.yaml
│   ├── MC_Scale100_Split82_InitOrd_ValueOnlySeq_Cot0_llama-3-8b-instruct_Epochs5_En.yaml
│   └── MC_Scale100_Split82_SeqeOrd_JsonSeq_Cot0_Qwen2.5-7B-Instruct_Epochs5_En.yaml
│
├── data/
│   ├── raw/
│   ├── processed_llm/
│   └── processed_tabular/
│
├── checkpoints/                                             // training checkpoints file
│   └── cross_entropy/
│
├── environments/                                            // environments file
│
├── base_models/                                             // LLMs base file
│
├── src/
│   ├── compare_base_model/
│   ├── common.py
│   ├── compare_model.py
│   ├── compare_test.py
│   ├── data_base_hacker_earth.py
│   ├── data_base_hacker_earth_onehot.py
│   ├── data_base_interface.py
│   ├── data_base_mathor_cup.py
│   ├── data_base_mathor_cup_onehot.py
│   ├── data_process.py
│   ├── generate_bash_script.py
│   ├── generate_cross_parameter_yaml.py
│   ├── llm_cross.py
│   ├── llm_response_analyser.py
│   └── llm_utils.py
│
├── templates/                                               // training template file
│
├── scripts/                                                 // batch scripts file
│   ├── predict.sh
│   └── train.sh
│
├── results/                                                 // test result file
│   └── metrics/
│
├── logs/
├── requirements.txt
└── llm_logistics.py                                         // main file