OpenViSTAI · Agents · Blockchain · Crypto
Submit a project
Menu
AI Models

DREAMERV3

DREAMERV3 is a Python and JAX reinforcement-learning implementation that learns a world model from experience and trains an actor-critic policy using imagined trajectories.

Explore the practical guide ↓

Independent OpenViST project profile, based on linked sources.

DREAMERV3 — OpenViST project profile in AI Models.
OpenViST project illustration. See the linked sources for product details.
Understand the project

Practical overview

When it fits

Use it for hands-on research and experimentation with model-based reinforcement learning in supported or adapted environments.

Who it is for

  • Reinforcement-learning researchers
  • Developers building self-hosted training experiments
  • Teams comparing model sizes, training ratios, and environment configurations

Documented uses

  • Train agents on documented Atari, Crafter, and Minecraft-related tasks.
  • Evaluate imagined trajectories before policy updates.
  • Run configurable experiments from the command line.
  • Collect terminal, JSONL, Scope, WandB, or TensorBoard training outputs.
Before you choose

Requirements & limits

The repository is public and open source under the MIT license. It is documented for Linux and macOS, with Docker instructions and Python 3.11+ requirements.

  • A working JAX installation is required.
  • The default execution target is GPU, with CPU and TPU selectable through --jax.platform.
  • Some environments require additional packages.
  • The debug configuration is intended for fast debugging and does not learn a good model.
  • The README documents a custom-environment configuration path but does not provide a complete custom-environment integration example.

Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.

Source-based practical guide

Set up and start a Crafter training run

Install the documented dependencies and launch training with the Crafter configuration.

Before you start

  • Linux or macOS
  • Python 3.11 or later
  • JAX
  • Repository dependencies installed from requirements.txt

Steps

  1. Install JAX.

  2. Install the other dependencies with pip install -U -r requirements.txt.

  3. Run the documented training command with a log directory, the crafter configuration, and a training ratio of 32.

  4. Review terminal metrics, JSONL output, or Scope summaries produced for the run.

Expected outcome

A configured Crafter training run writes experiment outputs to the selected log directory.

Guide limitations

  • The README notes that some environments require additional packages.
  • The guide documents starting a run, not verifying its learned performance.
  • Hardware-specific JAX installation details are not covered here.
Documentation 1 ↗
Source-based practical guide

Select the execution platform for training

Choose CPU, GPU, or TPU execution through the documented JAX platform option.

Before you start

  • A working JAX installation compatible with the selected hardware
  • A configured DreamerV3 training command

Steps

  1. Prepare the training command and configuration.

  2. Add --jax.platform cpu, --jax.platform gpu, or --jax.platform tpu.

  3. Run the training command with the selected platform.

Expected outcome

Training is instructed to use the selected CPU, GPU, or TPU platform.

Guide limitations

  • The README states that compatible hardware and JAX installation are required.
  • This does not document hardware setup or performance results.
Documentation 1 ↗
Evidence and transparency

Sources & review

Official repository ↗
Source record details
Source availability
Active
Sources checked
30 September 2026
Original website title
danijar/dreamerv3 · Official homepage title · retained in its source language
How we review sourcesSuggest a correction