spikefly/SemanticVLA-LAM

SemanticVLA · Latent Action Model (LAM)

🚧 Placeholder. The URL is stable; checkpoints will be uploaded incrementally per the release roadmap.

Trace-conditioned Latent Action Model (LAM) checkpoints for SemanticVLA. The LAM is a small VQ codebook trained to discretize short trace + future-action segments; its tokens are predicted as an auxiliary semantic head by the VLM during downstream VLA training.

Released variants

Variant	Training data	Selection metric	Headline metric	Status
libero/v5	LIBERO trace data	LIBERO downstream VLA SR	4-suite mean SR ≈ 0.984 (with VLA)	🔜
oxe-three-dataset/v4-step12k	BridgeData V2 + Fractal (RT-1) + BC-Z	per-dataset macro action-probe R²	macro R² = 0.0012 @ 12k	🔜
oxe-bridge-only/v4-step16k	BridgeData V2 only	Bridge action-probe R²	R² = 0.0167 @ 16k (global code usage 0.6875)	🔜

Planned layout

SemanticVLA-LAM/
├── libero/
│   └── v5/
│       ├── pytorch_model.pt
│       ├── config.yaml
│       └── model_card.md
├── oxe-three-dataset/
│   └── v4-step12k/
│       ├── pytorch_model.pt
│       ├── config.yaml
│       └── model_card.md
└── oxe-bridge-only/
    └── v4-step16k/
        ├── pytorch_model.pt
        ├── config.yaml
        └── model_card.md

Sibling SemanticVLA checkpoint repos

Repo	Purpose
🤗 SemanticVLA-LIBERO	LIBERO-finetuned VLA, consumes the libero/ LAM
🤗 SemanticVLA-Bridge	Bridge-finetuned VLA (for SimplerEnv WidowX), consumes the oxe-bridge-only/ LAM

Related resources

• Code: https://github.com/Fei-Ni/SemanticVLA_Offcial
• Datasets: https://hf.co/collections/spikefly/semanticvla-datasets
• Collection · Model Zoo: https://hf.co/collections/spikefly/semanticvla-model-zoo

How to load (placeholder API)

from huggingface_hub import hf_hub_download
import torch

ckpt = hf_hub_download(
    repo_id="spikefly/SemanticVLA-LAM",
    filename="oxe-bridge-only/v4-step16k/pytorch_model.pt",
)
state = torch.load(ckpt, map_location="cpu")
# loader will be released with the code repo

Citation

@misc{semanticvla2026,
  title  = {SemanticVLA: Semantic Vision-Language-Action Models with Trace-Conditioned Latent Action Tokens},
  author = {SemanticVLA Authors},
  year   = {2026},
  url    = {https://github.com/Fei-Ni/SemanticVLA_Offcial}
}

License

Released under the MIT License.