We're releasing HY-Embodied-0.5, a family of foundation models for real-world embodied agents. The 2...
- 推出2B和32B两个版本，分别面向边缘部署与复杂推理场景
- 采用MoT架构、隐式token和自进化后训练等关键技术
- 2B模型在22项基准中16项超越同类SOTA，32B接近前沿水平
It strengthens spatial-temporal perception and embodied reasoning for prediction, interaction, and planning. 🤖
The suite includes: 🔹 2B for edge https://t.co/r2nplyxMTf" / X
Post
Conversation
We're releasing HY-Embodied-0.5, a family of foundation models for real-world embodied agents. The 2B model is now open source. It strengthens spatial-temporal perception and embodied reasoning for prediction, interaction, and planning. !Image 2: 🤖 The suite includes: !Image 3: 🔹 2B for edge deployment !Image 4: 🔹 32B for complex reasoning Key innovations: !Image 5: 🔹 Mixture-of-Transformers (MoT) architecture for modality-specific computation !Image 6: 🔹 Latent tokens for improved perceptual representation !Image 7: 🔹 Self-evolving post-training !Image 8: 🔹 On-policy distillation from large to small models Across 22 benchmarks, the 2B model outperforms similarly sized SOTA systems on 16 tasks. The 32B model approaches frontier-level performance. !Image 9: 🔗 GitHub: github.com/Tencent-Hunyua!Image 10: 🤗 Hugging Face: huggingface.co/tencent/HY-Emb