Open Models Almanac
Methodspost-trainingsupervised fine-tuning

single mention · post-training

Joint fine-tuning across multiple embodiments

Joint fine-tuning across multiple embodiments adapts a policy to several embodiments at once by fine-tuning on demonstrations from diverse data sources.1 It fine-tunes on a mixture of downstream datasets covering distinct embodiments and hardware setups, aiming for performance that holds across all of them.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. adapted not only to a single novel embodiment but also to multiple embodiments simultaneously through joint finetuning on demonstrations from diverse data sources — X-VLA: Soft-Prompted Transformer as Scalable Cross-Embodiment Vision-Language-Action Model
  2. we conduct joint finetuning experiments on a mixture of downstream datasets, including Libero, BridgeData, and Calvin-ABC, which include two distinct embodiments and three hardware setups — X-VLA: Soft-Prompted Transformer as Scalable Cross-Embodiment Vision-Language-Action Model

Also called joint finetuning.

Filed under post-training › supervised fine-tuning