Reward Model Card for rynnvalue

RynnValue is a language-conditioned robotic value model built on Qwen3-VL. Given a trajectory video and a task description, it predicts the remaining time, in seconds, until the task is completed.

This reward model is inference-only in LeRobot: it can't be trained with lerobot-train.

See the full documentation at LeRobot Docs.


How to Get Started with the Reward Model

Load the reward model in Python

import torch

from lerobot.configs.rewards import RewardModelConfig
from lerobot.rewards import make_reward_model

config = RewardModelConfig.from_pretrained("lerobot/RynnValue-8B")
reward_model = make_reward_model(config)
reward_model.eval()


# `batch` is the output of this model's preprocessor.
with torch.inference_mode():

    remaining_time_s = reward_model.predict_remaining_time(batch).remaining_time_s
    success = reward_model.predict_success(batch).success


Model Details

  • License: apache-2.0
Downloads last month
140
Safetensors
Model size
10B params
Tensor type
BF16
ยท
Video Preview
loading

Model tree for lerobot/RynnValue-8B

Finetuned
(1)
this model