We’re about to launch a new real-time API for SAM3.1, that’s optimized for low latency, if folks are interested.
We’re seeing up to 3.5x speedup (lossless) over the native PyTorch implementation. ~40ms (ours) vs 150ms (eager-mode).
https://vlm.run/gateway/models/facebook-sam3.1
https://huggingface.co/posts/Xenova/448209562329557
We’re about to launch a new real-time API for SAM3.1, that’s optimized for low latency, if folks are interested.
We’re seeing up to 3.5x speedup (lossless) over the native PyTorch implementation. ~40ms (ours) vs 150ms (eager-mode).
https://vlm.run/gateway/models/facebook-sam3.1
https://huggingface.co/posts/Xenova/448209562329557