Open Models Almanac
Methodsdata curationtokenization

single source · data curation

MolmoAct2-FAST Tokenizer

MolmoAct2-FAST Tokenizer is an action tokenizer that converts continuous, embodiment-specific robot trajectories into compact discrete action tokens for a language model.1 It maps a one-second trajectory to tokens via a frequency-domain transform, quantization, and byte-pair encoding into a 2048-token action vocabulary.2

How this definition was written

Summarised by glm-5.3-flash from 1 source; every sentence is checked against the quote it cites.

  1. converts continuous, embodiment-specific robot trajectories into compact discrete action tokens — MolmoAct2 Action Reasoning Models for Real-World Deployment
  2. maps a one-second action trajectory into a compact sequence of discrete tokens by representing the trajectory with a frequency-domain transform, quantizing the resulting coefficients, and applying byte-pair encoding — MolmoAct2 Action Reasoning Models for Real-World Deployment

Filed under data curation › tokenization