LeRobot Refactors Qwen2.5-VL, Drops Internal Backup for Native Transformers
Decision Brief
What changedLeRobot replaced its Qwen2.5-VL implementation from an internally backed-up subclass to inheriting from the native Transformers library subclass.
Why it mattersUsing native Transformers subclass reduces maintenance overhead and helps robot projects keep up with upstream updates.
Who should careOpen-source model users
Affected stackQwenHugging Face
Source confidenceHigh · Official release / blog / repo
This commit refactors the Qwen2.5-VL implementation in the wall-x module, replacing the internally backed-up code with a subclass of the native Transformers library. Previously, LeRobot maintained its own copy of Qwen2.5-VL code; now it directly inherits from HuggingFace Transformers' Qwen2.5-VL class. This means new features or fixes for Qwen2.5-VL will flow into LeRobot automatically, eliminating manual syncs. For teams using LeRobot for robot vision-language model development, this change reduces code coupling and improves compatibility, but requires the Transformers version in the environment to support the required interfaces.
Summary basis: official / RSS sourceCompiled from the source scope noted above; the original remains authoritative.
Sources
- LeRobot（HuggingFace 机器人）
HuggingFace LeRobot — embodied-AI / robotics learning library; new models, datasets, and policies.
- LeRobot（HuggingFace 机器人）
留言
登入后即可留言，和其他 builder 交换实测心得。
还没有留言，抢头香。