bert-base-NER
by dslim
108M params · token-classification · 702 likes · 1.6M downloads
bert-base-NER is a 108M parameter model. At Q4 quantization it requires 0GB of VRAM. It runs comfortably on GeForce RTX 4090 (12121 tok/s), GeForce RTX 5090 (18178 tok/s), M4 Max 128GB (4435 tok/s).
Inference providers
| Provider | $/1M in | $/1M out | Throughput |
|---|---|---|---|
| HF Inference |
GPU compatibility
| GPU | VRAM | Q4 Decode | Verdict |
|---|---|---|---|
| GeForce RTX 4090 | 24GB | 12121 tok/s | comfortable |
| GeForce RTX 5090 | 32GB | 18178 tok/s | comfortable |
| M4 Max 128GB | 128GB | 4435 tok/s | comfortable |
| M4 Pro 48GB | 48GB | 2217 tok/s | comfortable |
| M4 Pro 24GB | 24GB | 2217 tok/s | comfortable |
| A100 PCIe 80 GB | 80GB | 22208 tok/s | comfortable |
| H100 SXM5 80 GB | 80GB | 44668 tok/s | comfortable |
| GeForce RTX 3090 | 24GB | 10717 tok/s | comfortable |
| Radeon RX 7900 XTX | 24GB | 8862 tok/s | comfortable |
| GeForce RTX 4080 | 16GB | 8602 tok/s | comfortable |