tapas-large-finetuned-wtq
by google
337M params · table-question-answering · 148 likes · 36.4k downloads
tapas-large-finetuned-wtq is a 337M parameter model. At Q4 quantization it requires 0GB of VRAM. It runs comfortably on GeForce RTX 4090 (3899 tok/s), GeForce RTX 5090 (5847 tok/s), M4 Max 128GB (1426 tok/s).
Inference providers
| Provider | $/1M in | $/1M out | Throughput |
|---|---|---|---|
| HF Inference |
GPU compatibility
| GPU | VRAM | Q4 Decode | Verdict |
|---|---|---|---|
| GeForce RTX 4090 | 24GB | 3899 tok/s | comfortable |
| GeForce RTX 5090 | 32GB | 5847 tok/s | comfortable |
| M4 Max 128GB | 128GB | 1426 tok/s | comfortable |
| M4 Pro 48GB | 48GB | 713 tok/s | comfortable |
| M4 Pro 24GB | 24GB | 713 tok/s | comfortable |
| A100 PCIe 80 GB | 80GB | 7143 tok/s | comfortable |
| H100 SXM5 80 GB | 80GB | 14368 tok/s | comfortable |
| GeForce RTX 3090 | 24GB | 3447 tok/s | comfortable |
| Radeon RX 7900 XTX | 24GB | 2850 tok/s | comfortable |
| GeForce RTX 4080 | 16GB | 2767 tok/s | comfortable |