ERNIE-4.5-VL-28B-A3B-PT
by baidu
29.4B params · image-text-to-text · 94 likes · 55.1k downloads
ERNIE-4.5-VL-28B-A3B-PT is a 29.4B parameter model. At Q4 quantization it requires 15GB of VRAM. It runs comfortably on GeForce RTX 4090 (44 tok/s), GeForce RTX 5090 (66 tok/s), A100 PCIe 80 GB (81 tok/s).
Inference providers
| Provider | $/1M in | $/1M out | Throughput |
|---|---|---|---|
| Novita | 77 tok/s |
GPU compatibility
| GPU | VRAM | Q4 Decode | Verdict |
|---|---|---|---|
| GeForce RTX 4090 | 24GB | 44 tok/s | comfortable |
| GeForce RTX 5090 | 32GB | 66 tok/s | comfortable |
| M4 Max 128GB | 128GB | 16 tok/s | tight |
| M4 Pro 48GB | 48GB | 8 tok/s | tight |
| M4 Pro 24GB | 24GB | 8 tok/s | tight |
| A100 PCIe 80 GB | 80GB | 81 tok/s | comfortable |
| H100 SXM5 80 GB | 80GB | 164 tok/s | comfortable |
| GeForce RTX 3090 | 24GB | 39 tok/s | comfortable |
| Radeon RX 7900 XTX | 24GB | 32 tok/s | comfortable |
| M4 Max 64GB | 64GB | 16 tok/s | tight |