|
Download README.md from MultiSTEVE-1s/steve1-training-data: direct link, hf CLI and curl.
- Browser
- Download file 2.42 kB
-
https://huggingface.co/datasets/MultiSTEVE-1s/steve1-training-data/resolve/main/README.md
- Command line
-
hf download hf://datasets/MultiSTEVE-1s/steve1-training-data/README.md
-
curl -L -o README.md https://huggingface.co/datasets/MultiSTEVE-1s/steve1-training-data/resolve/main/README.md
2.42 kB
| language: | |
| - en | |
| tags: | |
| - STEVE-1 | |
| - minecraft | |
| - minerl | |
| - mineclip | |
| pretty_name: STEVE-1 Training Data With MineCLIP Embeddings | |
| size_categories: | |
| - 10K--100K | |
| license: other | |
| license_name: multisteve-1-research-artefact-terms | |
| # STEVE-1 Training Data With MineCLIP Embeddings | |
| This dataset contains the MineCLIP-embedded training data used for the MultiSTEVE-1s model zoo. It supports reproducing STEVE-1-style fine-tuning without regenerating MineCLIP embeddings. | |
| ## Contents | |
| Top-level directories: | |
| - `dataset_contractor/`: OpenAI Contractor Dataset episodes converted for STEVE-1 training. | |
| - `dataset_mixed_agents/`: VPT-generated Minecraft trajectories collected for STEVE-1-style training. | |
| Each episode directory contains files such as: | |
| - `frames.mp4`: gameplay frames. | |
| - `actions.pkl`: action labels. | |
| - `embeds_attn.pkl`: MineCLIP embeddings used by the training pipeline. | |
| - `len.txt`: episode length. | |
| - `metadata.json`: episode metadata. | |
| ## Usage | |
| Use the supplementary code release and place the dataset under: | |
| ```text | |
| data/dataset_contractor/ | |
| data/dataset_mixed_agents/ | |
| ``` | |
| Then build samplings and train using the release scripts: | |
| ```text | |
| train/2_create_sampling.sh | |
| python -m scripts.train_zoo --group all | |
| ``` | |
| Full commands and environment details are in `README.md`, `docs/data_and_weights.md`, `docs/training.md`, and `docs/reproducibility.md` in the supplementary code release. | |
| ## Limitations | |
| The dataset is for Minecraft/MineRL policy training. It inherits the scope and limitations of the original VPT and STEVE-1 data pipelines. The dataset is large and intended for offline training and analysis, not interactive browsing. | |
| ## Licence And Terms | |
| This repository contains Minecraft-derived gameplay data, embeddings, actions, and metadata produced through OpenAI VPT, STEVE-1, MineCLIP/MineDojo, MineRL, and Minecraft-related workflows. This repository uses `license: other` under the MultiSTEVE-1 research artefact terms included in the repository `LICENSE` file. | |
| This repository does not grant rights to upstream STEVE-1, VPT, MineCLIP/MineDojo, MineRL, OpenAI contractor-demonstration, or Minecraft assets beyond their own licences and terms. We do not claim a blanket open-data licence over raw `frames.mp4`, action, or metadata files derived from contractor or Minecraft videos. It is not an official Minecraft product and is not approved by or associated with Mojang or Microsoft. | |