albert-base-v2
by albert
12M params · fill-mask · 140 likes · 462.6k downloads
albert-base-v2 is a 12M parameter model. At Q4 quantization it requires 0GB of VRAM. It runs comfortably on GeForce RTX 4090 (110873 tok/s), GeForce RTX 5090 (166268 tok/s), M4 Max 128GB (40573 tok/s).
Inference providers
| Provider | $/1M in | $/1M out | Throughput |
|---|---|---|---|
| HF Inference |
GPU compatibility
| GPU | VRAM | Q4 Decode | Verdict |
|---|---|---|---|
| GeForce RTX 4090 | 24GB | 110873 tok/s | comfortable |
| GeForce RTX 5090 | 32GB | 166268 tok/s | comfortable |
| M4 Max 128GB | 128GB | 40573 tok/s | comfortable |
| M4 Pro 48GB | 48GB | 20286 tok/s | comfortable |
| M4 Pro 24GB | 24GB | 20286 tok/s | comfortable |
| A100 PCIe 80 GB | 80GB | 203136 tok/s | comfortable |
| H100 SXM5 80 GB | 80GB | 408570 tok/s | comfortable |
| GeForce RTX 3090 | 24GB | 98029 tok/s | comfortable |
| Radeon RX 7900 XTX | 24GB | 81065 tok/s | comfortable |
| GeForce RTX 4080 | 16GB | 78687 tok/s | comfortable |