OLMo 2 Preview Post-trained Models Collection These model's tokenizer did not use HF's fast tokenizer, resulting in variations in how pre-tokenization was applied. Resolved in latest versions. • 6 items • Updated 21 days ago • 4