vit-base-patch16-224
by google
87M params · image-classification · 940 likes · 4.1M downloads
vit-base-patch16-224 is a 87M parameter model. At Q4 quantization it requires 0GB of VRAM. It runs comfortably on GeForce RTX 4090 (15167 tok/s), GeForce RTX 5090 (22745 tok/s), M4 Max 128GB (5550 tok/s).
Inference providers
| Provider | $/1M in | $/1M out | Throughput |
|---|---|---|---|
| HF Inference |
GPU compatibility
| GPU | VRAM | Q4 Decode | Verdict |
|---|---|---|---|
| GeForce RTX 4090 | 24GB | 15167 tok/s | comfortable |
| GeForce RTX 5090 | 32GB | 22745 tok/s | comfortable |
| M4 Max 128GB | 128GB | 5550 tok/s | comfortable |
| M4 Pro 48GB | 48GB | 2775 tok/s | comfortable |
| M4 Pro 24GB | 24GB | 2775 tok/s | comfortable |
| A100 PCIe 80 GB | 80GB | 27788 tok/s | comfortable |
| H100 SXM5 80 GB | 80GB | 55891 tok/s | comfortable |
| GeForce RTX 3090 | 24GB | 13410 tok/s | comfortable |
| Radeon RX 7900 XTX | 24GB | 11089 tok/s | comfortable |
| GeForce RTX 4080 | 16GB | 10764 tok/s | comfortable |