Back to Leaderboard
Parameters
44.5M
FLOPs
121.2G
Input Size
640px
License
Apache-2.0
Architecture
Type
one-stage
Backbone
NAS-Optimized
Neck
QSP
Head
Decoupled
Benchmark Results
Performance on COCO val2017 across different hardware configurations
| Hardware | mAP@50-95 | FPS | Latency | VRAM |
|---|---|---|---|---|
| NVIDIA A100 (TensorRT FP16) | 52.3% | 91.8 | 10.9ms | 1801 MB |
| NVIDIA T4 (TensorRT FP16) | 52.2% | 34.6 | 28.9ms | 1860 MB |
| CPU (ONNX Runtime) | 52.2% | 4.0 | 252.9ms | 1813 MB |
Speed Breakdown (A100 TensorRT)
End-to-end latency breakdown showing preprocessing, inference, and postprocessing times
1.3ms
7.2ms
2.4ms
Preprocess
Inference
Postprocess (NMS)
Usage with LibreYOLO
from libreyolo import YOLO
# Load model
model = YOLO.from_pretrained("https://huggingface.co/Libre-YOLO/yolo-nas-l")
# Run inference
results = model.predict("image.jpg")
# Process results
for box in results.boxes:
print(f"Class: {box.cls}, Confidence: {box.conf:.2f}")nas-optimizedhigh-accuracy
Related Models (yolo-nas)