Scientific Research Tokenized: NexaSci 10B-Token Pretraining Reservoir

NexaSci Scientific Research Tokenized is a large-scale text corpus for AI pretraining, containing 10 billion tokens of scientific research material. The dataset is maintained by AethronPhantom and was last updated in May 2026. It includes the current production reservoir and archived legacy builds.

Use Cases

• Pretrain large language models on the 10-billion-token scientific corpus.
• Fine-tune models for scientific tasks using the NexaMat controller fine-tuning pack.
• Benchmark model performance across different versions of the archived legacy reservoir builds.

Strengths

• Contains a 10-billion-token production reservoir for large-scale training.
• Includes both active training artifacts and archived legacy builds for comparison.

Limitations

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

Provenance

Source

AethronPhantom via Hugging Face.

Collection Method

Produced from the NexaSci 10B data-engineering campaign.

Freshness

Last updated 2026-05-15 04:44:26; freshness should be verified.