detr-resnet-50
by facebook
42M params · object-detection · 937 likes · 320.9k downloads
detr-resnet-50 is a 42M parameter model. At Q4 quantization it requires 0GB of VRAM. It runs comfortably on GeForce RTX 4090 (31538 tok/s), GeForce RTX 5090 (47296 tok/s), M4 Max 128GB (11541 tok/s).
Inference providers
| Provider | $/1M in | $/1M out | Throughput |
|---|---|---|---|
| HF Inference |
GPU compatibility
| GPU | VRAM | Q4 Decode | Verdict |
|---|---|---|---|
| GeForce RTX 4090 | 24GB | 31538 tok/s | comfortable |
| GeForce RTX 5090 | 32GB | 47296 tok/s | comfortable |
| M4 Max 128GB | 128GB | 11541 tok/s | comfortable |
| M4 Pro 48GB | 48GB | 5770 tok/s | comfortable |
| M4 Pro 24GB | 24GB | 5770 tok/s | comfortable |
| A100 PCIe 80 GB | 80GB | 57783 tok/s | comfortable |
| H100 SXM5 80 GB | 80GB | 116221 tok/s | comfortable |
| GeForce RTX 3090 | 24GB | 27885 tok/s | comfortable |
| Radeon RX 7900 XTX | 24GB | 23059 tok/s | comfortable |
| GeForce RTX 4080 | 16GB | 22383 tok/s | comfortable |