Liquid AI's LFM2.5-VL-3B-DSpark checkpoint is a companion to its image-and-text model. Despite the 3B name, the model card identifies the drafter itself as 279.5 million parameters and specifies the corresponding target model it must accompany.
The drafter proposes output tokenThe chunk of text a model reads and writes in — roughly three-quarters of a word — and the unit AI usage is billed in.Full definition → that the target checks. Under greedy decoding, which always selects the most likely next token, this verification preserves the target's original text. The checkpoint therefore provides a way to accelerate an existing model while leaving that model responsible for the accepted answer.
Terms in this piece · Glossary
token — The chunk of text a model reads and writes in — roughly three-quarters of a word — and the unit AI usage is billed in.
