Spaces:
Sleeping
Sleeping
File size: 5,497 Bytes
e9315b2 6e3f176 e9315b2 6e3f176 e9315b2 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 |
---
title: JSSP OpenEnv
emoji: ⏰
colorFrom: green
colorTo: purple
sdk: docker
pinned: false
---
<p align="center">
<img src="assets/jssp_openenv.png" alt="jssp_openenv" width="400">
</p>
<p align="center">
<a href="https://huggingface.co/spaces/Wauplin/jssp_openenv" style="font-size: 1.2em;">Try it live on Hugging Face Spaces</a>
</p>
## Job shop scheduling problem (JSSP)
The [Job Shop Scheduling Problem](https://en.wikipedia.org/wiki/Job-shop_scheduling) (JSSP) is a classic optimization problem in operations research. Given a set of jobs, each consisting of multiple operations that must be performed in a specific sequence, and a set of machines, the goal is to schedule the operations on machines to minimize the total completion time (makespan).
**Key constraints:**
- Each job consists of a sequence of operations that must be completed in order
- Each operation requires a specific machine for a given duration
- Each machine can process only one operation at a time
- Once started, an operation cannot be interrupted
This implementation uses the OpenEnv framework to create a reinforcement learning environment where an agent (policy) learns to make scheduling decisions at each time step.
> !TIP
> For now, we only implement and run the FT06 problem. It is a well-known problem in the literature with a known optimal solution.
> Goal for training is to run arbitrary random environments.
## OpenEnv
[OpenEnv](https://github.com/meta-pytorch/OpenEnv) is a framework from Meta PyTorch and Hugging Face for building reinforcement learning environments. It provides:
- A standardized interface for environments with `Action` and `Observation` models
- A web-based interface for interactive exploration of environments
- A client-server architecture for distributed training and evaluation
- Integration with LLM-based policies for solving complex problems
This project implements a JSSP environment using OpenEnv, allowing you to:
- Interact with the environment through a web interface
- Test different scheduling policies (FIFO, Max-Min, LLM-based)
- Train reinforcement learning agents to solve JSSP instances
## Project Architecture
The project follows a client-server architecture using the OpenEnv framework:
### Core Components
**Models** (`src/jssp_openenv/models.py`):
- `JSSPAction`: Represents scheduling actions (list of job IDs to schedule)
- `JSSPObservation`: Contains the current state (machine status, job status, remaining operations)
**Environment** (`src/jssp_openenv/server/jssp_environment.py`):
- `JSSPEnvironment`: The core simulation environment that:
- Manages job progress and machine states
- Validates actions and enforces constraints
- Advances simulation time using SimPy
- Returns observations and rewards
**Client** (`src/jssp_openenv/client.py`):
- `JSSPEnvClient`: HTTP client that communicates with the environment server
- Handles action serialization and observation parsing
**Policies** (`src/jssp_openenv/policy.py`):
- `JSSPEnvPolicy`: Abstract base class for scheduling policies
- `JSSPFifoPolicy`: First-In-First-Out scheduling (schedules jobs by ID order)
- `JSSPMaxMinPolicy`: Max-Min scheduling (prioritizes longest operations)
- `JSSPLLMPolicy`: LLM-based scheduling using OpenAI-compatible APIs
**Solver** (`src/jssp_openenv/solver.py`):
- `solve_jssp()`: Orchestrates the solving process by:
- Resetting the environment
- Iteratively applying policy actions
- Tracking scheduled events for visualization
- Returning makespan and event history
**Visualization** (`src/jssp_openenv/gantt.py`):
- Generates Gantt charts showing the schedule timeline
## How to use
### Install
Install the package and its dependencies:
```bash
pip install -e .
```
For development with additional tools (pytest, ruff, etc.):
```bash
pip install -e ".[dev]"
```
**Note:** For LLM-based policies, you'll need to set the `HF_TOKEN` environment variable with your Hugging Face API token:
```bash
export HF_TOKEN=your_token_here
```
### Run server
To play with the environment locally, run
```
python app.py
```
and go to http://0.0.0.0:8000/web.
### Run policy
**FIFO policy** (always run first available job):
```
python run.py fifo
```
**Max-Min policy** (always run longest job first):
```
python run.py maxmin
```
**LLM policy** (ask an LLM to solve the problem)
```
python run.py llm --model-id "openai/gpt-oss-20b:groq"
python run.py llm --model-id "openai/gpt-oss-120b:cerebras"
python run.py llm --model-id "Qwen/Qwen3-32B:groq"
```
### Check results
The solver will resolve the problem using the policy and then plot a gantt chart of the solution in the `./charts` folder.
Here is an example:

## Current Results
Results as of Nov. 7, 2024 on FT06 problem instance. *Note: Non-scientific results, only ran 1 episode per policy.*
| Policy | Makespan |
|--------|----------|
| **Optimal solution** | **55** |
| `openai/gpt-oss-20b:groq` | 61 |
| FIFO | 68 |
| `openai/gpt-oss-120b:cerebras` | 69 |
| `Qwen/Qwen3-32B:groq` | 69 |
| Max-Min | 77 |
## Run with docker
Build the Docker image:
```bash
docker build -t jssp-openenv .
```
Run the container:
```bash
docker run -p 7860:7860 jssp-openenv
```
The web interface will be available at http://localhost:7860/web.
## TODO
- [ ] run on other example environments (FT10, FT20)
- [ ] run on random environments
- [ ] run multiple policies and summarize results
- [ ] trainer |