⚡ Performance and Efficiency Benchmarks

This section reports the performance on NPU with FastFlowLM (FLM).

Note:

• Results are based on FastFlowLM v1.0.0.
• Under FLM’s default NPU power mode (Performance)
• Newer versions may deliver improved performance.
• SmolVLA is a Vision-Language-Action (robotics policy) model. It is measured as end-to-end latency per inference (camera images + instruction → action chunk), not in tokens per second.

Test System:

AMD Ryzen™ AI 9 370 (Strix Point) with 32 GB DRAM; performance is comparable to other Strix Point and Strix Halo Point systems.

🚀 Inference Latency (ms per inference, with different camera input counts)

📦 Package and Model Download

SmolVLA ships as a self-contained package. The model can be downloaded here:

• HuggingFace: FastFlowLM/smolvla
• ModelScope: amd/smolvla

For the model card and usage notes, see SmolVLA.