tinyroberta-squad2
by deepset
82M params · question-answering · 113 likes · 98.8k downloads
tinyroberta-squad2 is a 82M parameter model. At Q4 quantization it requires 0GB of VRAM. It runs comfortably on GeForce RTX 4090 (16104 tok/s), GeForce RTX 5090 (24150 tok/s), M4 Max 128GB (5893 tok/s).
Inference providers
| Provider | $/1M in | $/1M out | Throughput |
|---|---|---|---|
| HF Inference |
GPU compatibility
| GPU | VRAM | Q4 Decode | Verdict |
|---|---|---|---|
| GeForce RTX 4090 | 24GB | 16104 tok/s | comfortable |
| GeForce RTX 5090 | 32GB | 24150 tok/s | comfortable |
| M4 Max 128GB | 128GB | 5893 tok/s | comfortable |
| M4 Pro 48GB | 48GB | 2946 tok/s | comfortable |
| M4 Pro 24GB | 24GB | 2946 tok/s | comfortable |
| A100 PCIe 80 GB | 80GB | 29505 tok/s | comfortable |
| H100 SXM5 80 GB | 80GB | 59345 tok/s | comfortable |
| GeForce RTX 3090 | 24GB | 14238 tok/s | comfortable |
| Radeon RX 7900 XTX | 24GB | 11774 tok/s | comfortable |
| GeForce RTX 4080 | 16GB | 11429 tok/s | comfortable |