distilgpt2
by distilbert
88M params · text-generation · 616 likes · 2.2M downloads
distilgpt2 is a 88M parameter model. At Q4 quantization it requires 0GB of VRAM. It runs comfortably on GeForce RTX 4090 (14885 tok/s), GeForce RTX 5090 (22323 tok/s), M4 Max 128GB (5447 tok/s).
GPU compatibility
| GPU | VRAM | Q4 Decode | Verdict |
|---|---|---|---|
| GeForce RTX 4090 | 24GB | 14885 tok/s | comfortable |
| GeForce RTX 5090 | 32GB | 22323 tok/s | comfortable |
| M4 Max 128GB | 128GB | 5447 tok/s | comfortable |
| M4 Pro 48GB | 48GB | 2723 tok/s | comfortable |
| M4 Pro 24GB | 24GB | 2723 tok/s | comfortable |
| A100 PCIe 80 GB | 80GB | 27273 tok/s | comfortable |
| H100 SXM5 80 GB | 80GB | 54854 tok/s | comfortable |
| GeForce RTX 3090 | 24GB | 13161 tok/s | comfortable |
| Radeon RX 7900 XTX | 24GB | 10883 tok/s | comfortable |
| GeForce RTX 4080 | 16GB | 10564 tok/s | comfortable |