🧩 Model Card: SmolVLA

• Type: Vision-Language-Action
• Think: No
• Tool Calling Support: No
• Base Model: lerobot/smolvla_base
• Quantization: bf16
• Max Context Length: 48 tokens
• Max Camera Input: 3 images

📝 Note:

• SmolVLA is a robotics policy model that maps camera images and language instructions directly to robot actions — it does not run in FLM’s standard CLI or Server chat modes.
• For detailed usage instructions, please refer to:
  • HuggingFace: FastFlowLM/smolvla-NPU2
  • ModelScope: amd/smolvla
• For measured inference latency on Strix Point, see SmolVLA benchmarks.