ERNIE-4.5-21B-A3B-Thinking
by baidu
21.8B params · text-generation · 774 likes · 2.0k downloads
ERNIE-4.5-21B-A3B-Thinking is a 21.8B parameter model. At Q4 quantization it requires 11GB of VRAM. It runs comfortably on GeForce RTX 4090 (60 tok/s), GeForce RTX 5090 (90 tok/s), A100 PCIe 80 GB (110 tok/s).
GPU compatibility
| GPU | VRAM | Q4 Decode | Verdict |
|---|---|---|---|
| GeForce RTX 4090 | 24GB | 60 tok/s | comfortable |
| GeForce RTX 5090 | 32GB | 90 tok/s | comfortable |
| M4 Max 128GB | 128GB | 22 tok/s | tight |
| M4 Pro 48GB | 48GB | 11 tok/s | tight |
| M4 Pro 24GB | 24GB | 11 tok/s | tight |
| A100 PCIe 80 GB | 80GB | 110 tok/s | comfortable |
| H100 SXM5 80 GB | 80GB | 221 tok/s | comfortable |
| GeForce RTX 3090 | 24GB | 53 tok/s | comfortable |
| Radeon RX 7900 XTX | 24GB | 43 tok/s | comfortable |
| GeForce RTX 4080 | 16GB | 42 tok/s | comfortable |