UW/olmo-mix-1124-subset-p99

This subset contains documents sampled uniformly at random from allenai/olmo-mix-1124. The longest 1% of documents in the subset are then truncated to the 99th percentile in document lengths, due to some extreme outliers in document length. The train split is used for tokenizer training, and the eval split is used for evaluating tokenizer encoding efficiency.