Metadata-Version: 2.2
Name: agent-gpt-aws
Version: 0.9.0
Summary: AgentGPT CLI for training and inference on AWS SageMaker
Home-page: https://github.com/ccnets-team/agent-gpt
Author: JunHo Park
Author-email: junho@ccnets.org
License: Dual Licensed (AGENT GPT COMMERCIAL LICENSE or GNU GPLv3)
Keywords: agent gpt reinforcement-learning sagemaker
Classifier: Programming Language :: Python :: 3
Classifier: License :: OSI Approved :: GNU General Public License v3 (GPLv3)
Classifier: License :: Other/Proprietary License
Classifier: Operating System :: OS Independent
Requires-Python: >=3.8
Description-Content-Type: text/markdown
Requires-Dist: msgpack
Requires-Dist: uvicorn
Requires-Dist: fastapi
Requires-Dist: requests
Requires-Dist: gymnasium>=1.0.0
Requires-Dist: gymnasium[mujoco]
Requires-Dist: typer
Requires-Dist: websocket-client
Requires-Dist: pyyaml
Requires-Dist: boto3
Requires-Dist: sagemaker
Provides-Extra: mujoco
Requires-Dist: gymnasium[mujoco]; extra == "mujoco"
Provides-Extra: mlagents
Requires-Dist: mlagents_envs==0.30.0; extra == "mlagents"
Requires-Dist: protobuf==3.20.0; extra == "mlagents"
Dynamic: author
Dynamic: author-email
Dynamic: classifier
Dynamic: description
Dynamic: description-content-type
Dynamic: home-page
Dynamic: keywords
Dynamic: license
Dynamic: provides-extra
Dynamic: requires-dist
Dynamic: requires-python
Dynamic: summary

# AgentGPT: Remote Env Integrated Cloud RL Training

**W&B Humanoid-v5 Benchmark (via Internet):** [Weights & Biases Dashboard](https://wandb.ai/junhopark/agentgpt-beta)

![How AgentGPT Works](https://imgur.com/r4hGxqO.png)
---

## Overview

AgentGPT is a one-click, cloud-based platform for distributed reinforcement learning. It lets you easily host your environment simulators—either locally or in the cloud—and connect them to a central training job on AWS SageMaker. This enables efficient data collection and scalable multi-agent training using a GPT-based RL policy.

## Installation

```markdown
pip install agent-gpt-aws --upgrade
```

### Configuration

- **Config hyperparams & SageMaker:**
  ```bash
  agent-gpt config --batch_size 256
  agent-gpt config --role_arn arn:aws:iam::123456789012:role/AgentGPTSageMakerRole
  ```
- **List & Clear current configuration:**
  ```bash
  agent-gpt list
  agent-gpt clear
  ```

### Simulation

- **Run your environment (gym/unity/unreal, etc.) before training starts:**   
  ```bash
   agent-gpt simulate local
   agent-gpt simulate cloud
  ```

### Training & Inference

- **Train a gpt model on AWS:**
  ```bash
  agent-gpt train
  ```

- **Run agent gpt on AWS:**
  ```bash
  agent-gpt infer
  ```

## Key Features

- **Cloud & Local Hosting:** Quickly deploy environments (Gym/Unity) with a single command.
- **Parallel Training:** Connect multiple simulators to one AWS SageMaker trainer.
- **Real-Time Inference:** Serve a GPT-based RL policy for instant decision-making.
- **Cost-Optimized:** Minimize expenses by centralizing training while keeping simulations local if needed.
- **Scalable GPT Support:** Train Actor (policy) and Critic (value) GPT models together using reverse transitions.
