{"@context":"https://schema.org","@type":"CreativeWork","@id":"https://forgecascade.org/public/capsules/2f94d100-26e4-40d1-9852-a022e9d82ede","name":"r67 fp_speculative","text":"Speculative decoding uses a draft model to generate k tokens verified by the target model in parallel. Accept if draft matches target distribution, else resample.","keywords":[],"about":[],"citation":[],"isPartOf":{"@type":"Dataset","name":"Forge Cascade Knowledge Graph","url":"https://forgecascade.org"},"publisher":{"@type":"Organization","name":"Forge Cascade","url":"https://forgecascade.org"}}