[2026-07] 🎉 我们发布了Hy-Embodied-RxBrain-1.0——技术报告、官方推理代码及模型权重。RxBrain(Hy-Embodied-RxBrain-1.0)是一个统一的具身认知多模态基础模型——该单一模型将语言推理与视觉想象相结合,具备三大核心能力:
这些能力通过交错生成实现统一:在单个自回归序列中,RxBrain交替生成推理文本和流匹配的想象帧——通过学习到的<Image>令牌决定何时进行想象——因此,具身规划能够逐步将要做什么与世界应呈现的样子相结合。
flash-attn 必需)unified_mot所基于的hunyuan_vl_mot主干网络):pip install git+https://github.com/huggingface/transformers@9293856c419762ebf98fbe2bd9440f9ce7069f1a注意: 目前标准的
transformers发行版本尚未包含hunyuan_vl_mot;因此需要使用此固定提交版本。我们后续将把这些改进合并到 Transformers 主分支中。
git clone https://github.com/Tencent-Hunyuan/Hy-Embodied-RxBrain-1.0.git
cd Hy-Embodied-RxBrain-1.0
pip install -r requirements.txt| 组件 | 参数规模 | 来源 |
|---|---|---|
| Hy-Embodied-RxBrain-1.0 | ~6.2 B | 🤗 tencent/Hy-Embodied-RxBrain-1.0 |
FLUX VAE (ae.safetensors) | 83.8 M | 从 FLUX 发布版本获取 |
将权重下载到本地目录——加载器直接读取 checkpoint 文件,因此 --ckpt 必须是本地路径,不能是 Hub 仓库 ID:
pip install -U "huggingface_hub[cli]"
hf download tencent/Hy-Embodied-RxBrain-1.0 --local-dir ./Hy-Embodied-RxBrain-1.0VQA(理解)路径仅需要主权重。图像生成(T2I/世界模型推演/交错规划)还需要外部FLUX VAE ae.safetensors。
加载Transformers处理器以及本仓库中提供的UnifiedMoT类,然后运行理解(VQA)任务。请从仓库根目录运行,以便model包可导入,并将MODEL_PATH指向您的本地下载文件(参见模型下载)。
import torch
from transformers.models.hunyuan_vl_mot import HunYuanVLMoTProcessor
from model import UnifiedMoTForConditionalGeneration, maybe_init_generation_path
from vqa_inference import answer
MODEL_PATH = "./Hy-Embodied-RxBrain-1.0" # local checkpoint directory, not the Hub id
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
dtype = torch.bfloat16
# Load processor & model
processor = HunYuanVLMoTProcessor.from_pretrained(MODEL_PATH, trust_remote_code=True)
model = UnifiedMoTForConditionalGeneration.from_pretrained(MODEL_PATH, dtype=dtype)
maybe_init_generation_path(model, model_load_path=MODEL_PATH) # wires up the generation path
model.to(device).eval()
# Ask a question about an image
text = answer(
model, processor,
image_paths=["demo_cases/bridgev2_move_toy/input/obs_1.jpg"],
question="What objects are on the stovetop, and where is the green toy?",
device=device, dtype=dtype, max_new_tokens=256,
)
print(text)注意: RxBrain 使用自定义的交错式文本/图像解码循环,而非标准的
model.generateAPI。answer(...)辅助函数(位于vqa_inference.py中)为理解场景封装了该循环;图像生成和规划功能则分别有其下方的独立入口点。
以下任务同样提供即开即用的脚本:
纯自回归文本理解 — 无需 VAE / 流匹配。
python vqa_inference.py \
--ckpt ./Hy-Embodied-RxBrain-1.0 \
--images demo_cases/bridgev2_move_toy/input/obs_1.jpg \
--question "What objects are on the stovetop, and where is the green toy?" \
--max_new_tokens 256python text2image_inference.py \
--ckpt ./Hy-Embodied-RxBrain-1.0 --vae /path/to/ae.safetensors \
--prompt "a watercolor painting of a cat" \
--height 256 --width 256 --num_steps 25 --out out.png
# with classifier-free guidance
python text2image_inference.py \
--ckpt ./Hy-Embodied-RxBrain-1.0 --vae /path/to/ae.safetensors \
--prompt "a watercolor painting of a cat" \
--cfg_scale 5.0 --num_steps 50 --out out.pngpython multiframe_inference.py \
--ckpt ./Hy-Embodied-RxBrain-1.0 --vae /path/to/ae.safetensors \
--frames /path/to/obs.jpg --task "imagine the next frames" \
--num_frames 4 --num_steps 50 --out_dir multiframe_out对捆绑场景运行交错式规划。详情请参见 demo_cases/README.md。
CASE=umi_fold_sock
python interleave_inference.py \
--ckpt ./Hy-Embodied-RxBrain-1.0 --vae /path/to/ae.safetensors \
--frames demo_cases/$CASE/input/*.jpg \
--task "$(cat demo_cases/$CASE/prompt.txt)" \
--max_frames 5 --num_steps 50 --out_dir out_$CASERxBrain 在具身理解、空间推理以及想象/生成基准上进行了评估。有关详细指标和方法,请参阅我们的技术报告。