bert-base-cased
by google-bert
109M params · fill-mask · 350 likes · 3.3M downloads
bert-base-cased is a 109M parameter model. At Q4 quantization it requires 0GB of VRAM. It runs comfortably on GeForce RTX 4090 (12053 tok/s), GeForce RTX 5090 (18075 tok/s), M4 Max 128GB (4410 tok/s).
Inference providers
| Provider | $/1M in | $/1M out | Throughput |
|---|---|---|---|
| HF Inference |
GPU compatibility
| GPU | VRAM | Q4 Decode | Verdict |
|---|---|---|---|
| GeForce RTX 4090 | 24GB | 12053 tok/s | comfortable |
| GeForce RTX 5090 | 32GB | 18075 tok/s | comfortable |
| M4 Max 128GB | 128GB | 4410 tok/s | comfortable |
| M4 Pro 48GB | 48GB | 2205 tok/s | comfortable |
| M4 Pro 24GB | 24GB | 2205 tok/s | comfortable |
| A100 PCIe 80 GB | 80GB | 22083 tok/s | comfortable |
| H100 SXM5 80 GB | 80GB | 44416 tok/s | comfortable |
| GeForce RTX 3090 | 24GB | 10656 tok/s | comfortable |
| Radeon RX 7900 XTX | 24GB | 8812 tok/s | comfortable |
| GeForce RTX 4080 | 16GB | 8554 tok/s | comfortable |