Tmax 15K Open Instruct: RL Environment Instances for Model Training

Tmax 15K Open Instruct is a collection of approximately 15,000 reinforcement learning environment instances used to train the Tmax 9B and other Tmax models. The dataset was created by AllenAI and is formatted for use with their open-instruct fork. Details on its generation and composition are available in the associated research paper.

Use Cases

• Fine-tuning language models for instruction-following based on the described RL environment instances.
• Training or evaluating models on tasks generated through reinforcement learning procedures mentioned in the description.
• Researching the impact of specific RL-generated data on model performance, as referenced in the associated paper.

Strengths

• Approximately 15,000 data instances, providing a defined scale for training.
• Created by a known research organization, AllenAI, with an associated research paper for methodology.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

AllenAI

Collection Method

Generated as a collection of reinforcement learning environment instances; specific methodology detailed in the associated paper.

Freshness

Last updated 2026-06-23 03:28:43; freshness should be verified.