DeepSeek-V4-Pro
DeepSeek- Mixture of experts, 49B active
- MIT
- Released in 4-bit
- Estimated
deepseek-ai/DeepSeek-V4-Pro on Hugging Face
DeepSeek-V4-Pro with Glyd
| As released | Glyd | Change, % | ||
|---|---|---|---|---|
| Weights in GPU memory, GB4-bit weights counted as they are | 864.72 | 862.88est. | 0 | |
| 8×H100 nodes it takesWeights and an 8K-token KV cache | 2 | 2 | ||
| H100 80 GB GPUs it takes | 11 | 11 |
Which single GPU it fits
4-bit weights counted as they are.
| Memory | GPUs | As released | Glyd |
|---|---|---|---|
| 16 GB | RTX 4080, RTX 5080 | No | No |
| 24 GB | RTX 4090, RTX 3090, A10 | No | No |
| 32 GB | RTX 5090 | No | No |
| 48 GB | RTX A6000, L40S, RTX 6000 Ada | No | No |
| 80 GB | H100, A100 80 GB | No | No |
| 96 GB | RTX PRO 6000, GH200 | No | No |
| 141 GB | H200 | No | No |
Related models
Measured so far
Twenty-two open models measured, with quality wherever bf16 fits one GPU.