Tongyi-DeepResearch-30B-A3B
by Alibaba-NLP
30.5B params · text-generation · 803 likes · 14.0k downloads
Tongyi-DeepResearch-30B-A3B is a 30.5B parameter model. At Q4 quantization it requires 15GB of VRAM. It runs comfortably on GeForce RTX 4090 (43 tok/s), GeForce RTX 5090 (64 tok/s), A100 PCIe 80 GB (78 tok/s).
GPU compatibility
| GPU | VRAM | Q4 Decode | Verdict |
|---|---|---|---|
| GeForce RTX 4090 | 24GB | 43 tok/s | comfortable |
| GeForce RTX 5090 | 32GB | 64 tok/s | comfortable |
| M4 Max 128GB | 128GB | 15 tok/s | tight |
| M4 Pro 48GB | 48GB | 7 tok/s | tight |
| M4 Pro 24GB | 24GB | 7 tok/s | tight |
| A100 PCIe 80 GB | 80GB | 78 tok/s | comfortable |
| H100 SXM5 80 GB | 80GB | 158 tok/s | comfortable |
| GeForce RTX 3090 | 24GB | 38 tok/s | comfortable |
| Radeon RX 7900 XTX | 24GB | 31 tok/s | comfortable |
| M4 Max 64GB | 64GB | 15 tok/s | tight |