bert-large-uncased
by google-bert
336M params · fill-mask · 145 likes · 635.8k downloads
bert-large-uncased is a 336M parameter model. At Q4 quantization it requires 0GB of VRAM. It runs comfortably on GeForce RTX 4090 (3905 tok/s), GeForce RTX 5090 (5856 tok/s), M4 Max 128GB (1429 tok/s).
Inference providers
| Provider | $/1M in | $/1M out | Throughput |
|---|---|---|---|
| HF Inference |
GPU compatibility
| GPU | VRAM | Q4 Decode | Verdict |
|---|---|---|---|
| GeForce RTX 4090 | 24GB | 3905 tok/s | comfortable |
| GeForce RTX 5090 | 32GB | 5856 tok/s | comfortable |
| M4 Max 128GB | 128GB | 1429 tok/s | comfortable |
| M4 Pro 48GB | 48GB | 714 tok/s | comfortable |
| M4 Pro 24GB | 24GB | 714 tok/s | comfortable |
| A100 PCIe 80 GB | 80GB | 7154 tok/s | comfortable |
| H100 SXM5 80 GB | 80GB | 14390 tok/s | comfortable |
| GeForce RTX 3090 | 24GB | 3452 tok/s | comfortable |
| Radeon RX 7900 XTX | 24GB | 2855 tok/s | comfortable |
| GeForce RTX 4080 | 16GB | 2771 tok/s | comfortable |