rubert-tiny2
by cointegrated
29M params · sentence-similarity · 160 likes · 1.0M downloads
rubert-tiny2 is a 29M parameter model. At Q4 quantization it requires 0GB of VRAM. It runs comfortably on GeForce RTX 4090 (44692 tok/s), GeForce RTX 5090 (67021 tok/s), M4 Max 128GB (16354 tok/s).
Inference providers
| Provider | $/1M in | $/1M out | Throughput |
|---|---|---|---|
| HF Inference |
GPU compatibility
| GPU | VRAM | Q4 Decode | Verdict |
|---|---|---|---|
| GeForce RTX 4090 | 24GB | 44692 tok/s | comfortable |
| GeForce RTX 5090 | 32GB | 67021 tok/s | comfortable |
| M4 Max 128GB | 128GB | 16354 tok/s | comfortable |
| M4 Pro 48GB | 48GB | 8177 tok/s | comfortable |
| M4 Pro 24GB | 24GB | 8177 tok/s | comfortable |
| A100 PCIe 80 GB | 80GB | 81882 tok/s | comfortable |
| H100 SXM5 80 GB | 80GB | 164691 tok/s | comfortable |
| GeForce RTX 3090 | 24GB | 39514 tok/s | comfortable |
| Radeon RX 7900 XTX | 24GB | 32676 tok/s | comfortable |
| GeForce RTX 4080 | 16GB | 31718 tok/s | comfortable |