license: mit
tags:
- JSSP
- combinatorial_optimization
- llms
- supervised
- dataset
- llm_reasoning
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
dataset_info:
features:
- name: num_jobs
dtype: int64
- name: num_machines
dtype: int64
- name: prompt_jobs_first
dtype: string
- name: instruction
dtype: string
- name: prompt_machines_first
dtype: string
- name: input
dtype: string
- name: output
dtype: string
- name: path
dtype: string
- name: matrix
dtype: string
splits:
- name: train
num_bytes: 2278729467
num_examples: 129720
download_size: 752195130
dataset_size: 2278729467
Dataset Descriptions
Starjob introduces the first large-scale, supervised dataset (130,000 instances) specifically designed for training Large Language Models (LLMs) to solve the Job Shop Scheduling Problem (JSSP). Leveraging natural language representations of scheduling problems and solutions, Starjob enables fine-tuning of LLMs (Llama 8B, 4-bit quantized, trained with RsLoRA) for end-to-end scheduling. Our fine-tuned model not only generates feasible schedules, but also outperforms classic Priority Dispatching Rules (PDRs) and the state-of-the-art L2D neural approach, achieving up to 15% better results on standard benchmarks (Taillard, DMU). This work demonstrates the untapped potential of LLMs in combinatorial optimization and paves the way for interactive, explainable, and efficient scheduling solutions using natural language interfaces.
Jssp
Directory: jssp
File: starjob_no_absalute_path.json
Attributes:
num_jobsnum_machinesprompt_jobs_firstinstructionprompt_machines_firstinputoutputpathmatrixNumber of Items: 129720
Dataset Groupings
The datasets can be grouped based on their problem domains and attributes:
Other Problems
Datasets:
- Jssp Common attributes:
matrixoutputprompt_jobs_firstpathinstructionnum_jobsprompt_machines_firstnum_machinesinput
Statistics Summary
| Dataset | File | Number of Items | Attributes |
|---|---|---|---|
| Jssp | starjob_no_absalute_path.json | 129720 | num_jobs, num_machines, prompt_jobs_first, instruction, prompt_machines_first, input, output, path, matrix |